ai Trend Report
Dashboard へ戻る
3
High impact
5
Mid impact
367
Signal watch
なぜこのサイトを作ったのか
私たちそれぞれが個別にAIを使って情報収集し、同じような3行要約を作るたびに、世界中で膨大な電力と計算リソースが消費されています。 本プロジェクトは、あらかじめ広範な情報を取得・集約しておくことで、個別のAI実行回数を減らし、地球環境（GPU/TPU負荷）に配慮した効率的な情報収集を目指す実験的なダッシュボードです。
＠IT 全フォーラム 最新記事一覧
「光ファイバーはもう古い」「2035年宇宙の旅」 実運用へ向かうデータセンター新技術
・AI利用が加速する一方、データセンターは電力や土地、データの長期保存といった課題に直面している。そうした中で台頭してくる可能性がある、データセンター関連の新技術をまとめた。
ITmedia NEWS 最新記事一覧
キオクシアQ1、純利益が前年同期比4500％増 株式分割・自社株買いも
・キオクシアホールディングス（HD）が7月31日に公開した2027年3月期第1四半期（26年4月1日?6月30日）連結決算は、売上収益が1兆7671億1700万円（前年同期比415.5％増）、営業利益が1兆2700億1700万円（同2728.6％増）、純利益が8421億6500万円（同4506.6％増）と大幅な増収増益だった。AIサーバを巡るNAND需要の拡大が業績をけん引した。
Zennの「大規模言語モデル」のフィード
将棋エンジンなし、ツールありでClaude Fable 5に将棋を指させたら、ピヨ馬(R1220)にあと一歩で逆転負けしちゃったよ
・導入 claude-shogiは、将棋エンジンのヒントを一切使わず、Claude自身が局面解析ツールを頼りに考えて指す「Claude思考モード」を中心に据えた、Claude Code上で将棋を指すための環境だ。
・きっかけは、将棋YouTuberそらさんの動画「政府に規制された最強AIと将棋を指したらヤバすぎた」だった。指し手の符号だけをテキストで入力し、生成AI(Fable 5)と将棋を指す試みが紹介されていて、将棋専用に訓練されたAIではないにもかかわらず、序盤は正確な指し回しを見せていた。
・これを見て、合法手判定・詰み探索・候補手検証のような補助ツールをLLMに使わせれば、...
ITmedia NEWS 最新記事一覧
Amazon決算、AWS成長率36.7％で過去18四半期最速に 純利益はAnthropic投資益で大幅増
・Amazonの2026年4月?6月期決算は、売上高が前年同期比20％増の2006億600万ドル、純利益はAnthropic投資の評価益も寄与し約3.4倍の626億4700万ドルとなった。AWSの売上高は36.7％増と成長が加速。AI需要に応えるため、通期の設備投資予測を約2200億ドルへ上方修正した。
ITmedia NEWS 最新記事一覧
Apple、16％増収で第3四半期として過去最高 クックCEO最後の決算「将来をかつてないほど楽観」
・Appleの4?6月期決算は、売上高が前年同期比16％増の1094億1700万ドルで過去最高を記録した。iPhoneやMac、サービス部門が力強く牽引した一方、iPadは減収となった。ティム・クックCEOは今回が最後の決算会見となり、ジョン・ターナス次期CEOへの移行が順調に進んでいると語った。
ITmedia NEWS 最新記事一覧
ソニー、タムロン買収提案の狙いを説明 「イメージング事業の発展につながる」
・ソニーグループの陶琳CFOは7月31日の決算説明会で、傘下のソニーがレンズメーカーのタムロンに提案している完全子会社化について「両者が一緒になることで強みを生かせる」と狙いを語った。
ITmedia NEWS 最新記事一覧
ソニーの熊本工場、8月中旬には「地震前の稼働水準」に 早期復旧の理由を決算会見で明かす
・ソニーグループの陶琳CFOは7月31日の決算説明会で、令和8年熊本地震で生産を止めた熊本の半導体工場が8月中旬に地震発生前の稼働水準へ戻る見通しを示した。2016年の熊本地震では回復まで3カ月余りかかっており、早期復旧の理由を「被害の程度が違う」と説明し
ITmedia NEWS 最新記事一覧
一般消費者が「空調服」と書いたら商標権侵害？ 公式Xの注意喚起が波紋、弁理士の見解は
・ファン付きウェア「空調服」を展開する株式会社空調服が、公式Xで「空調服は登録商標」として一般名称の使用を呼びかけ、批判を受けて謝罪した。一連の投稿について、知的財産に詳しい栗原潔弁理士は「商標権者が呼びかけを行うことは賢明」との見解を示す。
cs.LG updates on arXiv.org
(Towards) Scalable Reliable Automated Evaluation with Large Language Models
・arXiv:2607.28282v1 Announce Type: cross Abstract: Evaluating the quality and relevance of textual outputs from Large Language Models (LLMs) remains challenging and resource-intensive.
・Existing automated metrics often fail to capture the complexity and variability inherent in LLM-generated outputs.
・Moreover, these metrics typically rely on explicit reference standards, limiting their use mostly to domains with objecti
＠IT 全フォーラム 最新記事一覧
「AlmaLinux 9.2」がセキュリティ認証取得、企業の“OS選定”にどう影響する？
・サイバートラストは「AlmaLinux 9.2」が国際標準規格「CC認証」を取得したと発表した。背景には企業が考慮すべきセキュリティリスクがあるという。
ITmedia NEWS 最新記事一覧
「ICチップなら安心」は本当か 地面師対策で問われる不動産取引の多重防御
・不動産の「ミステリー」を専門家がわかりやすく読み解き、AIをはじめITを活用した不動産の近未来を探る。バブル経済の崩壊後、低迷してきた日本の不動産価格が反転上昇し、内外からの投資が盛り上がる中、海外の先進事例なども交えて将来の不動産業界や価格をわかりやすく展望する。第4回は「地面師詐欺をITで防げるのか」を深堀りする。
＠IT 全フォーラム 最新記事一覧
「アイデア出しを抜いた」 生成AIなしで最も不安になる業務といえば？
・サイバーセキュリティクラウドが実施した調査で、上司よりも生成AIを参考にした経験を持つ人が半数に上るなど、職場でのAI依存が進んでいる実態が明らかになった。
Zennの「大規模言語モデル」のフィード
【LLM-as-a-Judge】少量のデータでジャッジを人手評価に近づける手法を比較する
・はじめに こんにちは、AI Shift AIチームの林崎です。
・LLMの出力品質をLLMに採点させるLLM-as-a-judgeを運用していて、「このジャッジ、本当に人間と同じ判断をしているのか？」と不安になったことはありませんか？ ジャッジを人間の評価に近づけるには人手の評価ラベルが必要ですが、実務で顧客や専門家からもらえるラベルは、数十件〜数百件になるケースが多いです。
・一方で、ジャッジを人手評価に近づける手法はFew-shot・評価基準の自動生成・プロンプト最適化など乱立しており、「手元のテストデータならどれを使うべきか」の実験的な比較は意外と見つかりません。
cs.LG updates on arXiv.org
$\beta$-OPSD: Deriving with Policy Optimization, Training with Self-Distillation
・arXiv:2607.28582v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) is a promising approach to improve reasoning language models, but it remains brittle in practice: making it work reliably often requires substantial engineering effort.
・We identify a structural source of this difficulty: vanilla OPSD is precisely the $\beta=1$ member of a broader policy-optimization family, where $\beta$ weights the KL
Zennの「大規模言語モデル」のフィード
2026-07-27 今日の技術トレンド
・本日は、LLM単体の進化よりもAIエージェントの実装・運用・安全性が主役でした。BigGoのWAIC報道や各社イベント記事が示すように、業界はエージェント中心へ移行中です。一方でReuters、Time、CNBCなどのOpenAI関連報道により、自律AIの事故対応、透明性、ガバナンスが最重要テーマとして浮上しました。Web開発ではNext.js/React/Vercelの直接ニュースは見当たらず、今日はPython基盤とAIツール群、特にuv・Ruff・Astral周辺の整備が実務上の注目点です。
・AI/LLM動向 今日のAI/LLM領域では、LLMそのものより“エージェント化...
cs.LG updates on arXiv.org
A Distributed Acoustic Sensing Dataset for Vessel Detection and Localization in Submarine Cable Protection
・arXiv:2607.28306v1 Announce Type: cross Abstract: Recent incidents of accidental damage and suspected sabotage to submarine telecommunication and power cables, particularly in the Baltic Sea, have underscored their vulnerability and the need for continuous monitoring solutions.
・Distributed acoustic sensing (DAS) applied to submarine optical-fiber cables enables wide-area monitoring of underwater acoustic activity.
cs.LG updates on arXiv.org
A Lightweight Foundation Model for Collider Physics with Multi-Domain Adaptation
・arXiv:2607.27501v1 Announce Type: new Abstract: We present a lightweight approach to foundation modeling (\textbf{NEXUS}) that leverages pre-trained learning from collider physics data towards out-of-domain tasks in other scientific datasets, using a fully connected autoencoder model with approximately 3 million parameters.
・The model pre-trains with no supervision over a large-scale collision dataset from the Large H
stat.ML updates on arXiv.org
A Mathematical Framework for Topological Causal Data Analysis
・arXiv:2607.28161v1 Announce Type: cross Abstract: Many modern outcomes, including images, point clouds, networks, and spatial fields, are structured objects for which \(Y^1-Y^0\) may be undefined or scientifically inadequate.
・We introduce \emph{Topological Causal Data Analysis} (TCDA), a framework separating the observation space, causal-model class, topological representation, and causal query.
・Topology does not def
cs.LG updates on arXiv.org
A Montage-Agnostic Encoder for Calibration-Light Cross-User Gesture Recognition from Surface Electromyography
・arXiv:2607.27565v1 Announce Type: new Abstract: Pattern-recognition control promises a myoelectric prosthesis that responds to many intended gestures rather than one or two, but the promise has stayed in the laboratory.
・A recogniser trained on one person rarely transfers to the next, and useful performance usually demands a fresh round of labelled calibration from the end user.
・A montage-agnostic encoder is introduce
cs.LG updates on arXiv.org
A novel k-means clustering approach using two distance measures for Gaussian data
・arXiv:2511.17823v2 Announce Type: replace Abstract: Clustering algorithms have long been the topic of research, representing the more popular side of unsupervised learning.
・Since clustering analysis is one of the best ways to find some clarity and structure within raw data, this paper explores a novel approach to k-means clustering.
・Here we present a k-means clustering algorithm that takes both the within cluster dis
cs.LG updates on arXiv.org
A Query-Efficient Stochastic Volume Rendering Framework for Time-Varying Implicit Neural Volumes
・arXiv:2607.28047v1 Announce Type: cross Abstract: Time-varying implicit neural representations (INRs) provide a compact representation of scientific volumes and, for modalities such as dynamic X-ray computed tomography (CT), are often the only practical way to represent the data.
・However, interactive volume rendering of INRs is challenging, as cheap memory lookups are replaced by expensive neural inferences, hinderin
Takara TLDR - Daily AI Papers
A Query-Efficient Stochastic Volume Rendering Framework for Time-Varying Implicit Neural Volumes
・Time-varying implicit neural representations (INRs) provide a compact representation of scientific volumes and, for modalities such as dynamic X-ray computed tomography (CT), are often the only practical way to represent the data.
・However, interactive volume rendering of INRs is challenging, as cheap memory lookups are replaced by expensive neural inferences, hindering the performance.
・Therefore, conventional volume
cs.LG updates on arXiv.org
Accelerating SGDM via Learning Rate and Batch Size Schedules: A Lyapunov-Based Analysis
・arXiv:2508.03105v3 Announce Type: replace Abstract: We analyze the convergence behavior of stochastic gradient descent with momentum (SGDM) under dynamic learning-rate and batch-size schedules by introducing a novel and simpler Lyapunov function.
・We extend the existing theoretical framework to cover three practical scheduling strategies commonly used in deep learning: a constant batch size with a decaying learning ra
stat.ML updates on arXiv.org
Adaptive Nystr\"om for Gaussian Process Regression
・arXiv:2607.27427v1 Announce Type: cross Abstract: Gaussian Process Regression (GPR) is a robust framework for uncertainty quantification, yet its $O(n^3)$ complexity limits its scalability.
・Low-rank Nystr\"om approximations can reduce this burden to $O(nm^2)$, but their accuracy depends heavily on the selection of landmark points.
・We propose an adaptive Nystr\"om approach that greedily selects landmarks to minimize t
cs.LG updates on arXiv.org
Adaptive Weighted LSSVM for Multi-View Classification
・arXiv:2512.02653v2 Announce Type: replace Abstract: Multi-view learning integrates diverse representations of the same instances and can improve performance when interactions across views are effectively exploited.
・Most existing kernel-based multi-view learning methods either rely on fusion techniques without explicitly enforcing a consensus or complementary collaboration across views, or use co-regularization-based
OpenAI News
Advancing responsible AI across Europe
・OpenAI shares how its safety, security, transparency, and provenance practices support responsible AI governance in Europe.
・The work will continue as the EU AI Act advances.
Takara TLDR - Daily AI Papers
Agentic Method for Deterministic Validation of Legacy Code Migration
・Migration of legacy COBOL programs to Java requires extensive testing to ensure correct functionality.
・This effort is often complicated by the lack of test data and the difficulty of validating all corner cases.
・In this paper we propose a novel agentic test-synthesis method, the "Locksmith Loop," which is initiated by preparing two runtime environments: the COBOL source and the generated Java target are each instrume
＠IT 全フォーラム 最新記事一覧
AIに「絶対するな」が通じない理由 「Claude Code」の意図しない動作を防ぐ7つのTIPS
・AIの意図しない挙動やプロンプトによる指示の「無視」を防ぐにはどうすべきなのか。Anthropicは「Claude Code」における7つの制御手法とそれぞれの実践的な使い分け、アンチパターンを解説した。
＠IT 全フォーラム 最新記事一覧
AIの台頭で見直しを迫られるマルチクラウド戦略
・AIの台頭により、企業のマルチクラウド戦略は見直しを迫られている。従来のベンダーロックイン回避や冗長性確保を目的とした防御的な活用から、戦略的な技術要件へと変化した。独自AI機能の違いやGPU不足などを踏まえ、ワークロードを最適なクラウドへの意図的な配置が重要だ。本稿では、AI主導のマルチクラウドを促進する4つの構造的要因と、I＆Oリーダーが対処すべき運用課題を具体的に紹介する。
Zennの「大規模言語モデル」のフィード
AIを分業させて「自動分析ツール」を簡単に作ろう
・データ分析をAIに丸投げしたい AIに作業を丸投げしたい。
・それも、単純作業じゃない。データ分析みたいな「頭を使う知的な作業」を！ AIワークフローなら、それができる。
・実際に、「自動データ分析AI」を作ってみた。
Zennの「大規模言語モデル」のフィード
AI体験記 vol.15 — ファイルの中に、AIへの命令が仕込まれていた
・前回（vol.14）の続きです。ここまで作ってきた土台が、悪意ある指示に耐えられるのかを考えた回です。
・技術的な正確さより、体験の正直さ優先。復習も兼ねています。
・vol.14までで、ハーネスはずいぶんレベルアップして頑丈になった。
cs.LG updates on arXiv.org
Albilich: Steerable Proof-State Orchestration for LLM-Based Mathematical Research with CAS Integration
・arXiv:2607.27705v1 Announce Type: cross Abstract: Large language models can contribute useful ideas to mathematical research, yet long-horizon proof attempts remain difficult to coordinate, evaluate, and reproduce.
・We present Albilich, an open-source agentic harness for autoresearch in mathematics that combines long-horizon reasoning, computer algebra systems (CAS), literature retrieval, and persistent SQLite-based c
cs.LG updates on arXiv.org
An analysis of binary isotonic regression: degrees of freedom and implications for calibration
・arXiv:2607.27301v1 Announce Type: cross Abstract: Isotonic regression is a canonical tool for estimating monotone functions and calibrating probabilistic predictors.
・We provide a fully sharp finite-sample characterization of its worst-case degrees of freedom on binary samples.
・Specifically, we identify the binary sequences that maximize the number of distinct fitted values produced by isotonic regression.
cs.LG updates on arXiv.org
APO: Unsupervised Atomic Policy Optimization for 3D Structure Prediction of Atomic Systems
・arXiv:2607.28553v1 Announce Type: new Abstract: Predicting the 3D structures of atomic systems is fundamental to advancing material science and drug discovery.
・While flow-matching models (, FlowDPO) have recently shown promise in this domain, their performance relies heavily on alignment with ground-truth coordinates via supervised preference learning.
・However, obtaining experimental labels for novel crystal phases o
cs.LG updates on arXiv.org
ARES: Anomaly Recognition Model For Edge Streams
・arXiv:2511.22078v2 Announce Type: replace Abstract: Many real-world scenarios involving streaming information can be represented as temporal graphs, where data flows through dynamic changes in edges over time.
・Anomaly detection in this context has the objective of identifying unusual temporal connections within the graph structure.
・Detecting edge anomalies in real time is crucial for mitigating potential risks.
cs.LG updates on arXiv.org
AskChem: Claim-Centered Infrastructure for Chemistry Literature Synthesis
・arXiv:2607.28618v1 Announce Type: cross Abstract: Chemistry literature synthesis often requires assembling specific findings scattered across many publications, yet existing literature-search systems primarily return ranked document lists.
・As a result, scientists and AI agents need to locate relevant information, verify their provenance, and assemble cross-paper answers manually.
・We present AskChem, a claim-centered
cs.LG updates on arXiv.org
AutoPref: Automatic Discovery of Task-Specific Preference Objectives for Neural Combinatorial Optimization
・arXiv:2607.27953v1 Announce Type: new Abstract: Combinatorial optimization problems (COPs) underpin many real-world decisions, but their exponentially large search spaces make high-quality solutions costly to obtain.
・Neural combinatorial optimization (NCO) learns fast construction policies, typically with reinforcement learning (RL), while preference-based NCO improves sample efficiency by learning from relative solu
cs.LG updates on arXiv.org
Averaged Evaluation Masks Capability Trade-Offs: Multi-Source Calibration for High-Sparsity LLM Pruning
・arXiv:2606.03328v3 Announce Type: replace Abstract: Calibration data are often treated as a minor implementation detail in post-training LLM pruning because averaged evaluations suggest only modest effects.
・We show that this conclusion is an averaging artifact: at 60\% SparseGPT sparsity, calibration strategies separated by only 2.85 points in averaged commonsense accuracy differ by 51.9 points in Code retention.
cs.LG updates on arXiv.org
Back from the Future: Key-Value Cache Management by Counter-Causal Surprise
・arXiv:2607.27600v1 Announce Type: new Abstract: Key-value (KV) cache management through compression and eviction strategies has emerged as an important research direction in recent years.
・Computational demands of large language models (LLMs) and their multi-modal variants during output generation can be partially alleviated by caching previous key and value calculations needed by subsequent scaled dot-product attenti
cs.LG updates on arXiv.org
Baikal: Structured Search for Deep Research over Data Lakes
・arXiv:2607.27726v1 Announce Type: cross Abstract: Deep research over data lakes requires an LLM agent to investigate evidence across thousands of heterogeneous tables and passages to synthesize a report.
・Existing methods perform iterative retrieval and generation, letting accumulated context determine what to investigate next, which can overexploit locally promising evidence and fail to cover distinct semantic region
stat.ML updates on arXiv.org
Bayesian Classification with Probit-link Split-and-merge Gaussian Process Prior in EEG-based Brain-Computer Interfaces
・arXiv:2605.30775v2 Announce Type: replace-cross Abstract: A Brain-Computer Interface (BCI) speller systems based on Event-Related Potentials (ERPs) enables users to select characters by detecting brain responses to visual stimuli, recorded through electroencephalogram (EEG).
・One challenge is to accurately identify target-related responses, such as the P300 component.
・However, existing methods tend to ignore feature s
cs.LG updates on arXiv.org
Benchmarking the Residual: What Long-Horizon Evaluations Add Beyond Matched Short-Task Performance
・arXiv:2607.27283v1 Announce Type: new Abstract: Long-horizon benchmarks often show that agents fail more as tasks become longer.
・This observation is useful for deployment, but it does not by itself explain why failure occurs.
・More stages create more opportunities for ordinary errors to compound; longer tasks may also contain harder individual decisions or become harder as conversation history, tool outputs, and envir
cs.LG updates on arXiv.org
Beyond Binary Rewards: A Comparative Study of Reward Design for Reinforcement Unlearning
・arXiv:2607.27968v1 Announce Type: new Abstract: Machine unlearning seeks to selectively remove specific knowledge from trained language models without full retraining, a growing necessity under privacy regulations such as GDPR and the EU AI Act.
・Recent work has reformulated unlearning as a Reinforcement Learning with Verifiable Rewards (RLVR) problem, where models are optimized against verifiable rewards computed dir
Takara TLDR - Daily AI Papers
Beyond Feeling Better: Capability-Sustaining Emotional Dialogue as a Longitudinal Research Paradigm
・Emotional dialogue research includes two influential strategy traditions.
・Empathetic dialogue prioritizes understanding a speaker's emotional experience.
・Emotional support conversation selects and sequences support for the seeker's current needs.
cs.LG updates on arXiv.org
Beyond Geometric Complementarity: Coherent Overlap in Sparse Mixture-of-Experts Routing
・arXiv:2607.28308v1 Announce Type: new Abstract: Sparse mixture-of-experts (MoE) language models route each token to multiple experts, suggesting a geometric account of their benefit: co-selected experts should contribute distinct representation directions.
・Existing evidence often conflates route coherence, candidate quality, and candidate-by-context interaction.
・We distinguish these quantities using an Expert Subspac
Takara TLDR - Daily AI Papers
Beyond Geometric Complementarity: Coherent Overlap in Sparse Mixture-of-Experts Routing
・Sparse mixture-of-experts (MoE) language models route each token to multiple experts, suggesting a geometric account of their benefit: co-selected experts should contribute distinct representation directions.
・Existing evidence often conflates route coherence, candidate quality, and candidate-by-context interaction.
・We distinguish these quantities using an Expert Subspace Separation Index (ESSI), matched-route residua
cs.LG updates on arXiv.org
Beyond KV Reconstruction: Functional Reconstruction for MLA Draft Models in Speculative Decoding
・arXiv:2607.27269v1 Announce Type: new Abstract: Multi-head latent attention (MLA) is increasingly important for long-context LLM inference because compact latent states replace the growing key-value (KV) cache and reduce decoding memory traffic.
・Yet most capable open checkpoints use multi-head or grouped-query attention (MHA/GQA), so conversion is needed to obtain MLA's cache efficiency without retraining from scratc
cs.LG updates on arXiv.org
Beyond the Best Teacher: Expanding and Compressing the Reasoning Solution Manifold
・arXiv:2607.27770v1 Announce Type: new Abstract: A single reinforcement-learning run can produce a strong reasoner yet an incomplete teacher: it often amplifies only a subset of the valid solution modes.
・We argue that reinforcement learning (RL)-trained policies should therefore be viewed as local probes of a multi-basin reasoning solution manifold, rather than as globally reliable supervisors.
・Based on this view, we
cs.LG updates on arXiv.org
Beyond the Bidirectional Promise: Re-evaluating the Robustness of Diffusion Language Models
・arXiv:2607.27386v1 Announce Type: cross Abstract: Diffusion Language Models (DLMs) offer a compelling alternative to autoregressive (AR) generation by enabling bidirectional context and iterative refinement.
・However, their reliability under natural input noise and adversarial attacks remains under-explored.
・To address this, we systematically evaluate DLM robustness and calibration against AR baselines, using two para
Takara TLDR - Daily AI Papers
BlindPSNR: A No-Reference Fidelity Predictor for Low-Light Image Enhancement
・Low-light image enhancement (LLIE) methods involve tunable parameters that are typically fixed, often leading to performance degradation when applied across scenes.
・Manually selecting the best configuration, however, can be time-consuming and not always practical.
・Peak signal-to-noise ratio (PSNR) is the natural fidelity criterion for automating parameter selection, yet it requires a ground-truth reference that is ty
cs.LG updates on arXiv.org
Bridging AI and Energy Forecasting: An Autonomous Workflow with Customized Toolkit
・arXiv:2307.07191v3 Announce Type: replace Abstract: Energy forecasting is crucial for the power grid, but fundamentally different from general time series analysis: it highly relies on covariates like meteorological factors, and its goals must align with actual power grid operations, such as risk assessment and system reliability.
・In order to bridge the huge gap between advanced machine learning forecasting models an
cs.LG updates on arXiv.org
Building a User Foundation Model for the Open Web
・arXiv:2607.28019v1 Announce Type: new Abstract: User foundation models have demonstrated strong results in e-commerce and social recommendation, but most industrial deployments assume environments where user identity is stable and persistent.
・Open-web real-time bidding (RTB) operates on a structurally different data distribution: user identity is fragmented and non-persistent across browsing sessions, and the availab
OpenAI News
Building abundant intelligence
・A full-stack approach to making advanced AI more capable, more affordable, and more widely useful.
cs.LG updates on arXiv.org
CACHE-UK: A Stability-Aware Memory Editor for Sequentially Updated Quantized LLMs in Finance
・arXiv:2607.28292v1 Announce Type: cross Abstract: Large Language Models (LLMs) deployed in dynamic financial environments face a critical challenge: maintaining factual accuracy as market conditions, regulations, and corporate facts change continuously.
・While 4-bit quantization enables efficient deployment, it severely limits the viability of sequential memory editing: existing methods undergo catastrophic performanc
Takara TLDR - Daily AI Papers
Can AI Follow In Einstein's Footsteps?
・AI is accelerating physics discovery, but perhaps away from Einstein-level theory building.
・To understand this gap, we must recognize a striking trend: while being very successful, the most visible AI contributions to physics discovery appear to mirror the historical development of physics, but in reverse.
・Human discovery in physics progressed, in broad strokes, from ancient pattern prediction, through phenomenologic
cs.LG updates on arXiv.org
Causal Discovery with Inverted Self-attention for Multivariate Time Series
・arXiv:2607.28212v1 Announce Type: cross Abstract: Causal discovery in multivariate time series data is challenging due to complex interactions, high dimensionality, and nonlinear dependencies among variables.
・Existing methods often struggle to capture these complexities, resulting in inaccurate causal structures.
・To address this issue, we propose a novel framework that leverages self-attention mechanisms within the t
cs.LG updates on arXiv.org
Certifying when decision-time information justifies adaptive experimentation
・arXiv:2607.27651v1 Announce Type: new Abstract: Adaptive laboratories choose measurements during experiments, yet most methods begin after adaptation is permitted.
・We introduce Opportunity-aware Policy Authorization for Laboratories (\OPAL{}), a framework that decides whether adaptation should be enabled at all.
・\OPAL{} uses a precommitted contract to require non-trivial adaptation, controlled target risk and positiv
cs.LG updates on arXiv.org
Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments
・arXiv:2607.28591v1 Announce Type: cross Abstract: Scaling coding agents requires a continuing supply of executable data for training, benchmarking, and continuous evaluation.
・Each task must couple a realistic software state with a specification, development tools, and reliable verification.
・To expand this supply, we present Change2Task, a system grounded in repository history that converts merged pull requests into v
Takara TLDR - Daily AI Papers
Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments
・Scaling coding agents requires a continuing supply of executable data for training, benchmarking, and continuous evaluation.
・Each task must couple a realistic software state with a specification, development tools, and reliable verification.
・To expand this supply, we present Change2Task, a system grounded in repository history that converts merged pull requests into verified tasks on healthy modern revisions of the s
cs.LG updates on arXiv.org
Chem World: A Large-Scale Benchmark and Physics-Informed Framework for Trustworthy Chemical Property Prediction
・arXiv:2607.28079v1 Announce Type: new Abstract: Chemical property prediction plays a critical role in accelerating scientific discovery in chemistry, materials science, and drug development.
・However, existing benchmarks often suffer from limited task diversity, fragmented datasets, and inconsistent evaluation protocols, making it challenging to systematically assess the reliability and generalization of AI models.
Takara TLDR - Daily AI Papers
ChronoMem: Version Control and Semantic Rollback for Large Language Model Agent Memory
・LLM agents increasingly rely on long-term memory to support multi-session interaction and personalization.
・However, existing agent memory systems are designed around forward-only evolution, continuously accumulating, consolidating, and overwriting knowledge, with no principled mechanism to inspect, version, or revert prior states.
・This makes agents brittle under corrections, concept drift, and memory corruption, part
cs.LG updates on arXiv.org
CLAM: Continuous Latent Action Models for Robot Learning from Unlabeled Demonstrations
・arXiv:2505.04999v2 Announce Type: replace-cross Abstract: Learning robot control policies from demonstrations typically requires action-labeled expert data, which is expensive to collect through teleoperation.
・We study a more practical setting in which expert demonstrations are available only as observation sequences without action labels, and only task-agnostic play data contains actions.
・We introduce continuous lat
cs.LG updates on arXiv.org
Class-Aware Reinforcement Learning for Counterfactual Explanation Generation
・arXiv:2607.27905v1 Announce Type: new Abstract: Counterfactual explanations (CFEs) enhance the interpretability of black-box models by generating alternative instances with adjusted feature values that achieve a contrastive outcome.
・Reinforcement learning (RL) offers a promising approach for CFE generation, enabling efficient exploration of counterfactual instances while ensuring control over key metrics like validit
Takara TLDR - Daily AI Papers
Class-Aware Reinforcement Learning for Counterfactual Explanation Generation
・Counterfactual explanations (CFEs) enhance the interpretability of black-box models by generating alternative instances with adjusted feature values that achieve a contrastive outcome.
・Reinforcement learning (RL) offers a promising approach for CFE generation, enabling efficient exploration of counterfactual instances while ensuring control over key metrics like validity, sparsity, and proximity.
・Previous studies hav
Zennの「大規模言語モデル」のフィード
Claude Code の Skill を eval 駆動で開発したら「効くのは検出力じゃなかった」
・結論 / TL;DR Claude Code の Skill（SKILL.md）を「なんとなく良さそう」で育てるのをやめて、eval 駆動に切り替えました。具体的には次の3点セットです。
・仕込みフィクスチャ — 既知の欠陥を1つだけ埋めた小さなプロジェクトを14ケース用意する with / without の A/B — 同じ入力を「Skill あり」と「素の Claude」の両方で走らせ、差分を測る アサーション採点 — ケースごとに合否判定できる観点リストで、各出力を採点する これで分かった一番大きなことは、Skill の価値は「検出力の上乗せ」ではなかった、という点で...
Zennの「大規模言語モデル」のフィード
Claude Codeを1回呼ぶと29,000トークン積まれる。プロンプトを削っても無駄だった
・claude -p をワーカーから大量に叩く構成（AIエージェントを常駐させる系）を作っていて、コストが気になったので実測しました。結果、自分が想定していた削りどころが全部ハズレでした。
・結論 claude -p は1回の呼び出しにつき固定で約29,000トークン積まれる。返事が6トークンでも変わらない プロンプトを削っても効果はほぼゼロ。CLAUDE.md（13KB）を読ませない空フォルダで測っても $0.070 → $0.068 効くのは起動回数とモデルだけ 15件を1回にまとめたら93%減（$0.90 → $0.067換算） ただしバッチ化は「溜める時間」とセットでないと...
ITmedia NEWS 最新記事一覧
Claudeも3組織に不正アクセス 実在システムを「演習の一部」と誤認 Anthropicが経緯を公表
・米Anthropicは7月30日（現地時間）、AIモデル「Claude」がサイバーセキュリティ評価の最中に3つの組織の本番システムへ不正アクセスしていたと発表した。隔離しているはずのテスト環境がインターネットにつながっており、Claudeは実在の企業を演習の標的と誤認したまま攻撃を進めていた。
cs.LG updates on arXiv.org
ClawTrack: Towards Trace-Level Evaluation and Improvement of Real-World Autonomous Agents
・arXiv:2607.28037v1 Announce Type: new Abstract: As LLM-based agents are deployed in complex, multi-step workflows, a critical evaluation gap has emerged: most existing benchmarks judge only final outcomes, unable to distinguish reliable reasoning from lucky success or attribute failures to specific process deficiencies, hindering attribution in long-horizon tasks.
・In this work, we present ClawTrack, a dual-assessment
cs.LG updates on arXiv.org
CLIP-Guided Backdoor Defense through Entropy-Based Poisoned Dataset Separation
・arXiv:2507.05113v3 Announce Type: replace-cross Abstract: Deep Neural Networks (DNNs) are susceptible to backdoor attacks, where adversaries poison training data to implant backdoor into the victim model.
・Current backdoor defenses on poisoned data often suffer from high computational costs or low effectiveness against advanced attacks like clean-label and clean-image backdoors.
・To address them, we introduce CLIP-Guid
cs.LG updates on arXiv.org
Comparison of a Parametric Physics-Informed Neural Network and a Tensorial Reduced-Order Model for the Shallow-Water Dam-Break Problem
・arXiv:2607.27433v1 Announce Type: cross Abstract: We develop two parametric data-driven reduced models: a physics-informed neural network (PINN) and a non-intrusive tensorial reduced-order model (TROM), and apply both approaches to the parametrized one-dimensional shallow-water dam-break problem.
・Both reduced models do not require time integration and learn a direct solution map from space, time, and dam-break parame
cs.LG updates on arXiv.org
Complementary Matrix-Gated QKAN Fast-Weight Programmers for Quantum Dynamics Forecasting
・arXiv:2607.27945v1 Announce Type: cross Abstract: Sequence models must decide what to write into memory and what to retain.
・In quantum and quantum-inspired sequence learning, nonlinear recurrent updates often require repeated circuit evaluations and sequential backpropagation through time, making long contexts costly.
・Gated fast-weight programmers (FWPs) based on quantum-inspired Kolmogorov-Arnold networks (QKANs) al
cs.LG updates on arXiv.org
Compliance2LoRA: On-Demand Safety Alignment on Arbitrary Policy Subsets via Hypernetwork-Generated LoRA Adapters
・arXiv:2607.27594v1 Announce Type: new Abstract: Post-training alignment in large reasoning models (LRMs) has significantly improved their adaptability to diverse safety compliance settings.
・However, as LRMs personalization for downstream users takes center stage, the demand for varying levels of policy compliance grows as different user-specific LRMs must adhere to distinct subsets of safety policies.
・Training a sepa
cs.LG updates on arXiv.org
Compression-Based Behavioral Similarity for Open-World Sybil Discovery on Ethereum
・arXiv:2607.27370v1 Announce Type: new Abstract: Sybil attackers are Blockchain actors that adopt the characteristics of regular users to exploit airdrops or influence governance.
・Current methods of Sybil actor detection include constructing graphs, which requires token transfers between examined wallets.
・Machine learning algorithms have been employed as well, but they treat the task as a closed-set classification pro
cs.LG updates on arXiv.org
Computer vision-based neural networks for radioisotope identification in urban environments
・arXiv:2607.00270v3 Announce Type: replace-cross Abstract: Algorithm development for radioisotope identification in mobile urban search scenarios face significant challenges from non-uniform backgrounds, momentary source encounters, and severe class imbalance between rare threat signatures and background measurements.
・We present a machine learning-based approach to this problem that converts list-mode gamma-ray data i
cs.LG updates on arXiv.org
Concurrent training methods for Kolmogorov-Arnold networks: Disjoint datasets and FPGA implementation
・arXiv:2512.18921v5 Announce Type: replace Abstract: The present paper introduces concurrency-driven enhancements to the training algorithm for the Kolmogorov-Arnold networks (KANs) that is based on the Newton-Kaczmarz (NK) method.
・Prior research shows that KANs trained using the NK-based approach outperform classical neural networks (multilayer perceptrons - MLPs) both in terms of accuracy and training time.
cs.LG updates on arXiv.org
Conformal Cascade: Distribution-Free Accuracy Guarantees for Multi-Tier LLM Inference
・arXiv:2607.25018v2 Announce Type: replace Abstract: Large language model (LLM) cascades reduce inference cost by routing easy queries to a small model and deferring hard queries to a larger one.
・Production cascades govern this deferral through a confidence threshold, but LLM confidence scores are miscalibrated, the threshold must be tuned per model pair and per domain, and no setting yields a formal bound on cascade
MIT News - Artificial intelligence
Connecting research to policy on Capitol Hill
・MIT students and postdocs discussed science funding and research with policymakers in Washington during the MIT Science Policy Initiative’s annual Congressional Visit Days.
cs.LG updates on arXiv.org
Context-Informed Ship Trajectory Prediction via Conditional Attention
・arXiv:2607.27418v1 Announce Type: new Abstract: Long-term ship trajectory prediction is a fundamental capability for maritime safety and autonomous navigation.
・While recent Transformer-based architectures have improved forecasting horizons, they predominantly rely on historical kinematic states, treating vessel motion as an isolated system.
・In reality, maritime navigation is profoundly modulated by extrinsic factors
cs.LG updates on arXiv.org
Continual Learning for VLMs: A Survey and Taxonomy Beyond Forgetting
・arXiv:2508.04227v3 Announce Type: replace-cross Abstract: Vision-language models (VLMs), spanning predictive architectures to generative Multimodal Large Language Models (MLLMs), have revolutionized artificial intelligence through powerful cross-modal alignment and zero-shot generalization.
・However, enabling them to learn continually from non-stationary data remains a major challenge, as their cross-modal alignment a
cs.LG updates on arXiv.org
Continual Learning with Vision-Language Models via Semantic-Geometry Preservation
・arXiv:2603.12055v3 Announce Type: replace-cross Abstract: Continual learning of pretrained vision-language models (VLMs) is prone to catastrophic forgetting, yet current approaches adapt to new tasks without explicitly preserving the cross-modal semantic geometry inherited from pretraining and previous stages, allowing new-task supervision to induce geometric distortion.
・We observe that the most pronounced drift tend
cs.LG updates on arXiv.org
Continuous-time reinforcement learning for optimal switching over multiple regimes
・arXiv:2512.04697v3 Announce Type: replace-cross Abstract: This paper studies the continuous-time reinforcement learning (RL) for optimal switching problems across multiple regimes.
・We consider a type of exploratory formulation under entropy regularization where the agent randomizes both the timing of switches and the selection of regimes through the generator matrix of an associated continuous-time finite-state Marko
cs.LG updates on arXiv.org
Contrastive Concept Importance: Explaining Pairwise Class Decisions Through Automatically Extracted Concept Representations
・arXiv:2607.27904v1 Announce Type: new Abstract: Concept-based explanations are a prevalent way to explain the decisions of complex black-box methods through semantically meaningful, human-interpretable concepts.
・To attribute the contribution of such concepts to a model's decisions, feature attribution methods are used to quantify how strongly each concept contributes to a model output.
・These attributions are typicall
cs.LG updates on arXiv.org
Contrastive Reinforced Policy Optimization via Privileged Self-Distillation
・arXiv:2607.28026v1 Announce Type: new Abstract: Recent advances in post-training Large Language Models (LLMs) increasingly rely on Reinforcement Learning with Verifiable Rewards (RLVR) or On-Policy Self-Distillation (OPSD).
・While OPSD provides dense, logit-level supervision, it inherently suffers from exposure bias due to the privileged information of the self-teacher.
・In multi-turn agentic settings, this leads to re
cs.LG updates on arXiv.org
Critic Architecture Matters: Dual vs. Unified Critics for Humanoid Loco-Manipulation
・arXiv:2606.11891v2 Announce Type: replace-cross Abstract: Multi-objective reinforcement learning for humanoid robots must coordinate locomotion and manipulation within a single policy.
・A natural design choice is whether to use a single (unified) critic that estimates the combined value of all objectives, or separate (dual) critics with disjoint reward signals.
・We compare the two on the Unitree G1 humanoid (23 active
cs.LG updates on arXiv.org
Critical attention scaling in long-context transformers
・arXiv:2510.05554v2 Announce Type: replace Abstract: As large language models scale to longer contexts, attention layers suffer from a fundamental pathology: attention scores collapse toward uniformity as context length $n$ increases, causing tokens to cluster excessively, a phenomenon known as rank-collapse.
・While $\textit{attention scaling}$ effectively addresses this deficiency by rescaling attention scores with a
cs.LG updates on arXiv.org
Cross-Embodiment Transfer via Behavior-Aligned Representations
・arXiv:2607.27549v1 Announce Type: cross Abstract: Recent progress in large-scale imitation learning for robot manipulation has been driven by leveraging datasets across a wide range of robot embodiments.
・However, achieving significant cross-embodiment transfer is often still challenging.
・In this work, we study the role of using behavior-aligned representations (e.g., object bounding boxes, language motions, end-effec
Takara TLDR - Daily AI Papers
CXR-Retrieve: Compositional Text-to-Image Retrieval in Chest Radiography
・Large chest radiography archives are difficult to search because most studies are paired only with free-text reports rather than structured clinical annotations.
・Vision-language models offer a natural interface for text-to-image retrieval, but current biomedical models are primarily optimized for report-to-image matching rather than for satisfying short clinical search queries.
・This creates an objective mismatch: a m
cs.LG updates on arXiv.org
Cybersecurity Detection Classification with Reasoning-enabled Language Models
・arXiv:2607.28460v1 Announce Type: new Abstract: A major issue in Security Operations Centers (SOCs) is alert fatigue, as the number of detections reported is more than staff can triage in a given day.
・Prior work prompts or fine-tunes large language models (LLMs) to emit a triage label directly, but does not train them to reason about whether a detection is a genuine threat.
・We train a chain-of-thought (CoT) reasoning
Takara TLDR - Daily AI Papers
Cybersecurity Detection Classification with Reasoning-enabled Language Models
・A major issue in Security Operations Centers (SOCs) is alert fatigue, as the number of detections reported is more than staff can triage in a given day.
・Prior work prompts or fine-tunes large language models (LLMs) to emit a triage label directly, but does not train them to reason about whether a detection is a genuine threat.
・We train a chain-of-thought (CoT) reasoning-enabled triage classifier on real, human-labele
MIT News - Artificial intelligence
Daniela Rus receives Bavarian Minister-President's High-Tech Prize
・Director of CSAIL and MIT professor honored for her contributions to robotics, artificial intelligence, and autonomous systems.
cs.LG updates on arXiv.org
DAS-PMVC: A Framework for Partial Multi-View Clustering via Dual Alignment and Structure Enhancement
・arXiv:2607.27761v1 Announce Type: new Abstract: In recent years, multi-view clustering has attracted widespread research interest.
・However, due to limitations in data collection devices, data across different views often suffer from misalignment, leading to the partial view alignment problem (PVAP).
・To mitigate the impact of view asymmetry and irrelevant samples, this paper proposes a framework for partial multi-view
stat.ML updates on arXiv.org
Debiased Machine Learning: Identification, Estimation, and Shape Constraints
・arXiv:2607.24472v2 Announce Type: replace-cross Abstract: We develop a general framework of identification and estimation for automatic debiased machine learning (DML) where the parameter of interest $\theta_0$ is identified by a moment condition involving a nuisance $\gamma_0$ that may be high dimensional.
・We establish conditions under which the Riesz representer $\alpha_0$, which is at the core of DML, is identifie
cs.LG updates on arXiv.org
Deep R Programming
・arXiv:2301.01188v5 Announce Type: replace-cross Abstract: Deep R Programming is a comprehensive and in-depth introductory course on one of the most popular languages for data science.
・It equips ambitious students, professionals, and researchers with the knowledge and skills to become independent users of this potent environment so that they can tackle any problem related to data wrangling and analytics, numerical com
cs.LG updates on arXiv.org
Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation
・arXiv:2606.13657v3 Announce Type: replace Abstract: On-policy distillation (OPD) has recently become a prominent post-training recipe by combining two desirable ingredients: on-policy student-generated trajectories and dense token-level teacher supervision.
・Yet how this hybrid training regime shapes a model remains poorly understood.
・We characterize the sparsity and geometry of OPD parameter updates across several la
cs.LG updates on arXiv.org
DIPHINE: Diffusion-based $\Phi$-ID Neural Estimator
・arXiv:2606.18997v2 Announce Type: replace Abstract: Uncovering the true informational architecture of real-world complex systems requires disentangling how their components uniquely store, redundantly share, and synergistically integrate information over time.
・Integrated Information Decomposition ($\Phi$ID) is a framework for decomposing the information dynamics of multivariate systems into sixteen non-overlapping at
OpenAI News
Disrupting a Criminal Scam Operation
・OpenAI disrupted a Cambodia-based scam operation using ChatGPT to support investment, romance, gambling, and impersonation schemes.
cs.LG updates on arXiv.org
Dissecting Federated-Graph Aggregation under Domain Shift: Importance-Aware Aggregation via Empirical Analysis
・arXiv:2509.18171v5 Announce Type: replace Abstract: Federated graph learning (FGL) trains a shared graph model across clients whose local graphs differ in node features, labels, and connectivity while keeping raw graph data decentralized.
・Although graph-domain shifts across clients can severely degrade the global model, existing FGL approaches for graph-domain shift mainly adapt local representations, propagation, or
cs.LG updates on arXiv.org
Distributions In, Distributions Out: The Case for Soft-Label Training
・arXiv:2511.14117v2 Announce Type: replace Abstract: Supervised classifiers output a distribution over classes but are typically trained against a single label obtained by collapsing multiple annotators into a majority vote.
・On tasks where annotator disagreement reflects genuine ambiguity -- natural language inference, politeness, visually ambiguous categorization -- this collapse discards information and forces model
cs.LG updates on arXiv.org
Divergence Decoding: Training-Free Capability Fusion
・arXiv:2607.27248v1 Announce Type: cross Abstract: While large language models excel in reasoning, these generalists often lack knowledge for specialized scientific domains.
・Conversely, domain models~(specialists), while knowledgeable, suffer from specialization side-effects including diminished logic and reduced robustness.To address this dilemma, we introduce Divergence Decoding, a training-free framework for capabi
cs.LG updates on arXiv.org
DoTime: A Synthetic Benchmark Generator for Interventional and Counterfactual Time Series
・arXiv:2607.27263v1 Announce Type: new Abstract: Most benchmarks for causal inference over time series are observational, small, or domain-specific, leaving interventional and counterfactual estimation under-served exactly where it matters most, such as in healthcare, policy evaluation, and climate science.
・We introduce \textbf{DoTime}, an open, scalable, and theoretically grounded generator of multivariate temporal s
cs.LG updates on arXiv.org
Doubly Robust Functional Representation Learning for Longitudinal Causal Inference with Irregular Histories
・arXiv:2607.28567v1 Announce Type: cross Abstract: Longitudinal causal studies often record histories as irregular functional fragments: laboratory values, physiologic signals, sensor streams, and image-derived summaries measured at unequal and informative times.
・Standard doubly robust estimators usually require scalar summaries, whereas sequence learners optimize prediction losses that need not stabilize the efficien
Takara TLDR - Daily AI Papers
Drawing-Recode: Annotation Grounding for Parametric CAD Code Generation from Raster 2D CAD Drawings
・Recovering Parametric CAD sequences from raster-format 2D Computer-Aided Design (CAD) drawings accumulated prior to digital transformation is important for part reproduction and manufacturing process automation.
・However, existing studies either process only vector drawings or are limited to specific domains, and fail to explicitly connect dimensional annotations to geometric information, limiting their use of dimensi
cs.LG updates on arXiv.org
Driving up Inference Energy on SNNs: Per-Sample and Universal Sponge Attacks
・arXiv:2607.27990v1 Announce Type: cross Abstract: Spiking Neural Networks (SNNs) communicate through sparse binary spike events rather than dense activations, enabling energy-efficient inference on neuromorphic hardware and motivating their use in always-on, battery-powered edge systems.
・We show that this same efficiency advantage creates a distinct security risk: sponge attacks can increase inference-time spike acti
cs.LG updates on arXiv.org
DS@GT ARC at ImageCLEFmedical 2026: Architectural Diversity for Concept Detection and Foundation-Model Scaling for Caption Prediction in Medical Image Analysis
・arXiv:2607.27763v1 Announce Type: cross Abstract: We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions.
・For Task 1, our primary submission was
Takara TLDR - Daily AI Papers
DS@GT ARC at ImageCLEFmedical 2026: Architectural Diversity for Concept Detection and Foundation-Model Scaling for Caption Prediction in Medical Image Analysis
・We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions.
・For Task 1, our primary submission was a three-way late-fusion ensemble of ConvNeXt-V2, B
cs.LG updates on arXiv.org
Dynamic Spectral Filtering for Temporal Graph Learning: Learning Evolving Propagation Operators
・arXiv:2607.27891v1 Announce Type: cross Abstract: Temporal graph learning is commonly organized around the evolution of node states or the encoding of interaction histories.
・We study an underexplored, operator-centric question: should the graph propagation mechanism itself evolve over time?
・We introduce Dynamic Spectral Filtering (DSF), which represents propagation at snapshot t by a Chebyshev polynomial filter with
cs.LG updates on arXiv.org
Dynamically Scaled Activation Steering
・arXiv:2512.03661v2 Announce Type: replace Abstract: Activation steering has emerged as a powerful method for guiding the behavior of generative models towards desired outcomes such as toxicity mitigation.
・However, most existing methods apply interventions uniformly across all inputs, degrading model performance when steering is unnecessary.
・We introduce Dynamically Scaled Activation Steering (DSAS), a method-agnostic
cs.LG updates on arXiv.org
ECG-InterpBench: Benchmarking the Interpretability of ECG Foundation Models with Matched-Scale Sparse Autoencoders
・arXiv:2607.27404v1 Announce Type: new Abstract: Existing benchmarks for electrocardiogram foundation models primarily evaluate downstream predictive performance, providing limited insight into whether their internal representations can be faithfully decomposed, clinically interpreted, or reproduced across independent analyses.
・We introduce ECG-InterpBench, a benchmark designed to systematically evaluate the interpret
cs.LG updates on arXiv.org
Echoverse: Deep, Evolving Environments for Training Computer-Use Agents at Scale
・arXiv:2607.28074v1 Announce Type: cross Abstract: Computer-use agents learn from what their actions change, so training one needs applications it can act on, break and reset.
・The applications that matter most are login-gated and stateful, so synthetic environments stand in for them.
・Recent pipelines generate such environments in bulk, which moves the bottleneck from how many exist to what is inside each one.
cs.LG updates on arXiv.org
Efficient LLMs with AMP: Attention Heads and MLP Pruning
・arXiv:2504.21174v2 Announce Type: replace Abstract: Deep learning drives a new wave in computing systems and triggers the automation of increasingly complex problems.
・In particular, Large Language Models (LLMs) have significantly advanced cognitive tasks, often matching or even surpassing human-level performance.
・However, their extensive parameters result in high computational costs and slow inference, posing challen
cs.LG updates on arXiv.org
EMBL AI Librarian: Life-Sciences Knowledge Layer for AI Agents
・arXiv:2607.28229v1 Announce Type: cross Abstract: The web is increasingly accessed by AI agents rather than humans.
・Every agent needs knowledge, especially in the life-sciences, where agentic pipelines are growing fast.
・Access to the literature is a crucial part of that need, and resources such as Europe PMC, with over 40M indexed records, are widely used to meet it.
cs.LG updates on arXiv.org
Emulating Cosmic Structure Formation with a Lagrangian Neural Cellular Automaton
・arXiv:2607.27320v1 Announce Type: cross Abstract: Field-level inference of cosmological initial conditions from galaxy surveys requires a forward model that is simultaneously accurate in the non-linear regime, computationally efficient, and fully differentiable.
・Traditional N-body simulations are accurate but computationally prohibitive for iterative inference, while approximate solvers like Lagrangian Perturbation T
cs.LG updates on arXiv.org
Encryption-Compatible Clustered Federated Learning via Distributed Expectation-Maximization over Metadata
・arXiv:2607.28338v1 Announce Type: new Abstract: Clustered Federated Learning (CFL) addresses data heterogeneity in federated settings by grouping clients with similar data distributions to enable effective training.
・Existing methods face a trade-off between privacy preservation, communication cost, and computational efficiency.
・We formalize this as the CFL trilemma, according to which improving two of these dimension
cs.LG updates on arXiv.org
Enhancing Irregular Time Series Forecasting with Continuous-Time Modeling Framework
・arXiv:2607.28035v1 Announce Type: new Abstract: Irregular multivariate time series are widely encountered in applications such as healthcare monitoring, human activity recognition, and environmental sensing.
・Their core challenges stem from asynchronous observations, non-uniform sampling intervals, and the fact that temporal patterns themselves carry critical dynamic information.
・Existing approaches either rely on dis
stat.ML updates on arXiv.org
Entropy-Smooth Convex Optimization Cannot Be Accelerated
・arXiv:2607.27476v1 Announce Type: cross Abstract: We prove an $\Omega(L/T)$ lower bound for the convergence rate of minimization in the class of functions that are convex and $L$-smooth relative to negative entropy on the standard $d$-simplex, valid for every first-order method when $d = \Omega(T^2)$.
・In particular, this shows that mirror descent is optimal up to a logarithmic factor in this class.
・This may be surpri
cs.LG updates on arXiv.org
Epistemic diversity across language models mitigates knowledge collapse
・arXiv:2512.15011v3 Announce Type: replace Abstract: Artificial intelligence (AI) increasingly generates the very content used to train future AI systems.
・This feedback loop can degrade model quality, reduce informational diversity, and ultimately drive knowledge collapse, i.e.
・a degradation to a narrow and inaccurate set of ideas.
cs.LG updates on arXiv.org
Error Analysis of Neural-Network-Based Engression
・arXiv:2607.27723v1 Announce Type: cross Abstract: Engression (Shen and Meinshausen, 2024) learns a conditional distribution by fitting a generative model $Y = f(X,\varepsilon)$ under the energy score, a strictly proper scoring rule.
・We provide a theoretical error analysis of engression implemented with deep neural networks.
・We decompose the excess risk into three components: the approximation error, the stochastic er
cs.LG updates on arXiv.org
Evaluation Protocols and Cross-Subject Generalization in EEG Emotion Recognition
・arXiv:2607.27655v1 Announce Type: new Abstract: Reported accuracy in electroencephalography (EEG) emotion recognition depends on the complete evaluation procedure, not only the classifier.
・We separate the target quantity, development procedure, and reporting rule, then use one archived dynamical graph convolutional neural network (DGCNN) pathway on SEED and SEED-IV as an illustrative case.
・In a protocol-matched subje
cs.LG updates on arXiv.org
Event-Structured Physics-Informed Neural Networks for Differentiable Critical Clearing Boundaries
・arXiv:2607.27681v1 Announce Type: new Abstract: Transient-stability assessment determines whether a power system can recover after a disturbance and is therefore essential to preventing generator trips and cascading outages.
・A key metric is the critical clearing time (CCT), which specifies the maximum time available to clear a fault before synchronism is lost.
・Reliable CCT estimation is challenging because complicate
cs.LG updates on arXiv.org
EvoCause: LLM-Guided Evolution of Causal Graphs for Root Cause Analysis
・arXiv:2607.27290v1 Announce Type: new Abstract: Modern telecommunication, cloud, and microservice systems emit correlated alarm cascades when components fail.
・Root cause analysis (RCA) aims to identify the small set of alarms that initiate each cascade.
・A common approach learns a causal graph from observational logs and predicts all zero-in-degree alarms in each incident-induced subgraph.
cs.LG updates on arXiv.org
Exact Action Values Are Not Enough: Rollout-Verified Reinforcement Fine-Tuning of a Reasoning Model for Multi-Zone VAV Control
・arXiv:2607.27914v1 Announce Type: new Abstract: Multi-zone variable-air-volume control must balance thermal comfort, indoor air quality, and electricity use across several continuous actuators.
・Model predictive control and reinforcement learning are widely studied, but deployment typically requires building-specific modeling or training, limiting scalability.
・We first test whether a frontier reasoning model (an LLM t
cs.LG updates on arXiv.org
Expanding Data-Agnostic Pivotal Instances Selection Models with Proximity Trees and Ensemble Learning
・arXiv:2607.27522v1 Announce Type: new Abstract: As decision-making processes grow more complex, machine learning tools have become essential for tackling business and societal challenges.
・However, many existing methods rely on decision-making procedures that are difficult to interpret.
・Since humans naturally make decisions by comparing new cases with a few representative examples, we aim to design an approach that se
cs.LG updates on arXiv.org
Expected Survival-Time Bounds for Robust Optimization Over Time under Isotropic Gaussian Dynamics
・arXiv:2607.27280v1 Announce Type: cross Abstract: Robust Optimization Over Time (ROOT) is a recent branch of evolutionary dynamic optimization that seeks solutions capable of remaining effective across multiple consecutive environments.
・Unlike the traditional track-the-moving-optimum (TMO) paradigm, which reoptimizes after every environmental change, ROOT explicitly values persistence.
・Although the field has grown co
cs.LG updates on arXiv.org
Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation
・arXiv:2607.27372v1 Announce Type: new Abstract: The deep learning revolution, kicked off by AlexNet, taught us that end-to-end training beats decomposing a problem into hand-designed stages.
・Generative modeling, however, has remained the exception-despite generative models being remarkably capable, they are still not trained end-to-end.
・This is because, at its core, generative modeling is about handling distributions
cs.LG updates on arXiv.org
FADEx: Feature Attribution and Distortion-based Explanation of Dimensionality Reduction
・arXiv:2607.27463v1 Announce Type: new Abstract: Dimensionality Reduction (DR) is a fundamental tool for high-dimensional data exploration, reducing the complexity of latent spaces of machine learning models, and assisting in the explanation of complex opaque models.
・However, non-linear DR techniques often function as opaque transformations themselves, making it challenging to understand how individual features influe
cs.LG updates on arXiv.org
Fairness Pruning: Locating Demographic Bias in GLU-MLP Layers via Differential Activations
・arXiv:2607.28319v1 Announce Type: cross Abstract: This work presents Fairness Pruning, a lightweight structural intervention method designed for the management and future mitigation of demographic bias in large language models (LLMs).
・As a foundational empirical validation of this method, this work focuses on causal bias localization.
・Using minimally contrastive prompt pairs and inference-time activation capture, the
Takara TLDR - Daily AI Papers
Fairness Pruning: Locating Demographic Bias in GLU-MLP Layers via Differential Activations
・This work presents Fairness Pruning, a lightweight structural intervention method designed for the management and future mitigation of demographic bias in large language models (LLMs).
・As a foundational empirical validation of this method, this work focuses on causal bias localization.
・Using minimally contrastive prompt pairs and inference-time activation capture, the method identifies neurons that react differential
cs.LG updates on arXiv.org
FeatFix: Reuse What You Verify through Local Exact-Feature Correction for Faster Cached Diffusion Inference
・arXiv:2607.27842v1 Announce Type: cross Abstract: Diffusion models are widely used to generate high-quality images and videos, but their iterative denoising process remains computationally intensive.
・A growing class of training-free accelerators reduces this cost by reusing cached intermediate features or forecasting future ones.
・To control draft drift, these methods sometimes compute an exact block feature for verif
cs.LG updates on arXiv.org
FedOGL: Combating Catastrophic Forgetting in Federated Open-World Multimodal Graph Learning
・arXiv:2607.27665v1 Announce Type: new Abstract: Federated graph learning enables collaborative training over decentralized graph data without sharing raw graph information.
・As such risks evolve, clients must learn emerging classes from private multimodal graph streams, retain historical categories, and reject samples outside the known class space.
・In this setting, clients must learn emerging classes from private mult
Takara TLDR - Daily AI Papers
Fidelity Is Not Safety: Gently-Compressed LLMs Pass Every Data-Free Quality Guard Yet Invent Procedure Steps in Agentic Execution
・Practitioners accept a compressed language model once it clears a stack of data-cheap quality guards: perplexity within a small factor of the original, downstream accuracy (for example MMLU) inside a confidence interval, and data-free output-fidelity signals that compare the compressed and original network's internal representations under random probe inputs.
・This stack has a blind spot.
・Across three model families,
cs.LG updates on arXiv.org
Filling the Pareto-Optimal Front for Affordance Segmentation on Embedded Devices Using RGB-D Cameras
・arXiv:2607.28293v1 Announce Type: cross Abstract: While depth sensors have the potential to complement RGB data for affordance segmentation in wearable robots, their usage seems to remain underexplored.
・The paper proposes two approaches: a reformulated version of hardware-aware neural architecture search, endowed with a newly designed search space to integrate depth (D) information into small-sized deep networks, and
cs.LG updates on arXiv.org
FinSMART: Financial Sentiment Analysis for Algorithmic Trading through Market-Aligned Reinforcement Learning
・arXiv:2607.28127v1 Announce Type: cross Abstract: Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs).
・However, existing approaches remain confined to a market-agnostic, supervised learning paradigm that relies on limited, static and human-annotated datasets, and thus are incapable of adapting to evolving market conditio
Takara TLDR - Daily AI Papers
FiRE: Enhancing MLLMs with Fine-Grained Context Learning for Complex Image Retrieval
・Due to their strong generalizable multimodal processing and reasoning capabilities, Multimodal Large Language Models (MLLMs) have demonstrated significant potential as universal image retrievers, effectively addressing diverse real-world image retrieval tasks.
・Nevertheless, pioneering studies, while promising, overlook the potential of fine-grained context modeling and disentangled fine-tuning objectives in enhancing
cs.LG updates on arXiv.org
First-order Constrained Trilevel Optimization Over Distributed Networks for Robust Coreset Selection
・arXiv:2607.27632v1 Announce Type: new Abstract: With the rapid advancement of the Internet of Things (IoT), massive amounts of data are generated across distributed edge networks.
・Training models on full data incurs significant computational overhead and storage bottlenecks, rendering coreset selection a critical paradigm.
・Furthermore, given the privacy-sensitive nature of local data and the escalating demand for mod
cs.LG updates on arXiv.org
Flat Score, Amplified Failures: How the Error Budget Masks Damage in Quantized LLM Agents
・arXiv:2607.27275v1 Announce Type: new Abstract: Post-training quantization to 4-bit weights is widely reported to be nearly lossless.
・We test this claim for multi-turn, tool-calling agents, where it now matters most.
・On $\tau^2$-bench, across two open-weight model families in dense and MoE variants and two domains (eight cells, 456 episodes each, at 16-, 8-, and 4-bit weights), quantization indeed looks free on the s
cs.LG updates on arXiv.org
Flux-OPD: On-Policy Distillation with Evolving Contexts
・arXiv:2607.28022v1 Announce Type: new Abstract: Large language model training in open-ended domains lacks verifiable rewards, making task preferences difficult to formalize as effective supervision.
・Contexts can convey such preferences, yet provide little additional supervision once distilled into the student, motivating contexts that evolve with student performance.
・However, directly using evolving contexts as in-tr
Takara TLDR - Daily AI Papers
Flux-OPD: On-Policy Distillation with Evolving Contexts
・Large language model training in open-ended domains lacks verifiable rewards, making task preferences difficult to formalize as effective supervision.
・Contexts can convey such preferences, yet provide little additional supervision once distilled into the student, motivating contexts that evolve with student performance.
・However, directly using evolving contexts as in-training supervision results in an unstable distil
cs.LG updates on arXiv.org
FORGE: Fused On-Register Gradient Elimination for Memory-Efficient LLM Training
・arXiv:2606.22932v2 Announce Type: replace Abstract: Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets the optimizer read it back.
・This two-phase schedule sets the memory ceiling of modern training: at the seam between the phases, every layer's gradient is live at once.
・We argue that this materialized gradient is an artifact of how differentiation is staged, not a qua
cs.LG updates on arXiv.org
Foundation-Model Earth Representations Enable Regional-Scale Forest Aboveground Biomass Monitoring Across the Northeastern United States
・arXiv:2607.27217v1 Announce Type: cross Abstract: Forest aboveground biomass (AGB) is a critical indicator of ecosystem productivity and terrestrial carbon storage, yet regional carbon monitoring remains constrained by the sparse spatial and temporal availability of field inventories and airborne structural measurements.
・Recent Earth observation foundation models provide globally consistent geospatial representations
cs.LG updates on arXiv.org
From Expert Reduction to Behavioral Divergence: Tracing Numerical State through Sparse MoE Inference
・arXiv:2607.28097v1 Announce Type: new Abstract: Mathematically equivalent expert-reduction orders can produce observably different sparse-MoE executions.
・We isolate this effect in native DeepSeek-V4-Flash by freezing local MoE state and varying only aggregation semantics.
・Four schemes separate operand representation from accumulator precision.
Takara TLDR - Daily AI Papers
From Minds to Models: The Intersection of Psychology and LLM Behaviours
・Large language models (LLMs) are often compared with the human mind because their decision-making is complex, non-linear and difficult to interpret.
・Psychological methods developed to investigate unobservable mental processes may therefore help examine LLM behaviour, particularly in government and healthcare.
・Building on prompt-based adaptations of the Implicit Association Test, this study tested whether ChatGPT prod
Takara TLDR - Daily AI Papers
From Single- to Cross-Document: Benchmarking Multi-Granularity Event Analysis of Large Language Models
・Event analysis is an essential and fundamental direction of information extraction, involving various event-centric tasks at different granularity of documents.
・While large language models (LLMs) have preliminarily achieved promising performance in part of these tasks individually, their capability in event analysis still lacks comprehensive understanding due to restricted document granularity, task designs, and data
Takara TLDR - Daily AI Papers
From Textual Requirements to Microservice Architectures - A Comprehensive Evaluation of LLM-Based Design Synthesis
・Microservice architectures have become dominant for modernizing monolithic systems, yet identifying appropriate services remains challenging and largely manual.
・Existing decomposition approaches are predominantly code-centric, limiting applicability in early design stages where only textual requirements are available.
・Despite advances in Large Language Models (LLMs), limited empirical evidence exists on their ability
cs.LG updates on arXiv.org
Fully Inductive Cardinality Estimation
・arXiv:2607.28311v1 Announce Type: cross Abstract: Query optimization of Basic Graph Patterns (BGP) SPARQL queries over Knowledge Graphs (KG) requires accurate cardinality estimation.
・Recently published learned estimators outperform statistics- and sampling-based approaches, but share a limitation preventing their adoption in real-world triplestores: they are transductive and require retraining when the underlying gra
cs.LG updates on arXiv.org
FunL2O: LLM-Guided Feature Function Design for Learning to Optimize
・arXiv:2607.27389v1 Announce Type: new Abstract: Learning-to-optimize (L2O) methods accelerate repeated optimization by training models to predict solutions, warm starts, branching decisions, or other forms of solver guidance.
・A critical yet largely overlooked component of these pipelines is the feature function that maps problem instances to inputs for machine learning models.
・Existing L2O methods typically rely on h
cs.LG updates on arXiv.org
Generalization and Trade-off in Adversarial Training: An RKHS Perspective via Kernel Integral Operators
・arXiv:2607.27995v1 Announce Type: cross Abstract: Adversarial training has emerged as a powerful approach for protecting models against adversarial attacks in a broad range of real-world applications.
・In this paper, we study adversarial training in the reproducing kernel Hilbert space (RKHS) framework through the associated kernel integral operator.
・We first derive source-uniform generalization error bounds for the R
cs.LG updates on arXiv.org
Generalization Bounds on Optimal Control for Transformer Training and Wasserstein Distributional Robustness
・arXiv:2607.27975v1 Announce Type: new Abstract: We derive finite-sample generalization bounds for Transformers trained with dynamic programming recursions.
・Building on the doubly lifted, measure-valued formulation of Transformer dynamics, we view data sets as probability laws on pairs of empirical input-output measures, allowing us to interpret the training problem as a finite-horizon Markovian control problem.
Takara TLDR - Daily AI Papers
Generalization Bounds on Optimal Control for Transformer Training and Wasserstein Distributional Robustness
・We derive finite-sample generalization bounds for Transformers trained with dynamic programming recursions.
・Building on the doubly lifted, measure-valued formulation of Transformer dynamics, we view data sets as probability laws on pairs of empirical input-output measures, allowing us to interpret the training problem as a finite-horizon Markovian control problem.
・We then analyze a quantized model, derived by quantiz
cs.LG updates on arXiv.org
Good Rankers, Bad Objectives: Bilinear Contrastive Critics under Expressive Policy Search
・arXiv:2607.27422v1 Announce Type: new Abstract: Good action rankings do not make a contrastive critic safe to maximize.
・These critics increasingly act as value-like objectives for best-of-$K$ selection, planning, and critic-guided generation.
・Unbounded bilinear scores can let large embedding norms inflate off-support values, but cosine bounding does not remove the failure.
cs.LG updates on arXiv.org
Gradient-free Task-Conditioned Retrieval for On-Device In-Context Learning
・arXiv:2607.27766v1 Announce Type: cross Abstract: On-device in-context learning (ICL) relies on pre-inference retrieval to select demonstrations for useful context before downstream model inference.
・This retrieval must exploit task-specific information while operating over local memories under limited computation, memory, and data-exposure budgets.
・We propose Conditional Retrieval Alignment (CoRA), a gradient-free fr
cs.LG updates on arXiv.org
Graph Neural Multilevel Preconditioners for Iterative Solvers
・arXiv:2607.28456v1 Announce Type: cross Abstract: Solving large, sparse linear systems is a core task in scientific computing, and efficient iterative solvers rely critically on effective and robust preconditioning.
・While classical methods such as algebraic multigrid (AMG) are highly scalable, their robustness can degrade on indefinite or nonsymmetric systems where heuristics originally developed for elliptic PDEs ar
cs.LG updates on arXiv.org
Graph Neural Network Force Fields for Spin Dynamics in Metallic Magnets
・arXiv:2607.28537v1 Announce Type: cross Abstract: Metallic magnets exhibit complex spin dynamics governed by electronically generated interactions.
・Predictive simulations of such dynamics typically require repeated solutions of an underlying electronic problem throughout the time evolution, creating a major computational bottleneck.
・Here we introduce a graph neural network (GNN) magnetic force-field framework that le
cs.LG updates on arXiv.org
Group-Reflective Self-Distillation for Agentic Reinforcement Learning
・arXiv:2607.28076v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) is effective for training large language model agents.
・However, terminal rewards provide only coarse trajectory-level supervision, leaving successful behaviors, recurring mistakes, and incidental choices entangled in the same outcome signal.
・Existing agentic self-distillation methods enrich sparse supervision with
cs.LG updates on arXiv.org
GVR-Coder: A Visual-Feedback Framework for Structured SVG Generation in Complex Document and Meeting Scenarios
・arXiv:2607.28073v1 Announce Type: new Abstract: In demanding professional environments and meeting review scenarios, lengthy text often imposes a high cognitive load.
・To facilitate efficient information communication, transforming verbose text into logically clear diagrams is essential.
・Scalable Vector Graphics (SVG) provide an effective representation for this purpose due to their editability and resolution independ
cs.LG updates on arXiv.org
GyRot: Leveraging Hidden Synergy between Rotation and Fine-grained Group Quantization for Low-bit LLM Inference
・arXiv:2607.27694v1 Announce Type: cross Abstract: Low-bit quantization is essential for efficient LLM inference, and both rotation and fine-grained group quantization have shown individual promise.
・However, their combination often leads to accuracy degradation or hardware overhead due to a mismatch between the global nature of rotation and the localized behavior of group scaling.
・We propose GyRot, a quantization fram
cs.LG updates on arXiv.org
HARGO: Heterogeneity-Aware Reward-Guided Optimization for RL Post-Training of LLMs on HPC Tasks
・arXiv:2607.28301v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) can equip large language models (LLMs) with domain knowledge for high-performance computing (HPC) tasks such as data race detection and benchmark question answering.
・However, knowledge alone does not guarantee task-appropriate behavior: the same SFT model that correctly classifies 88.65\% of C/C++ data race samples produces verbose, imprecis
cs.LG updates on arXiv.org
Harnessing the Potential of Optimizing Data Mixtures via Bayesian Domain Reweighting
・arXiv:2607.27928v1 Announce Type: new Abstract: The performance of Large Language Models (LLMs) is fundamentally influenced by the distributional composition of multi-domain pre-training data.
・While manual heuristics were prevalent in early models, they increasingly fail to capture the intricate synergies between domains as data complexity grows.
・To overcome the issue, a dominant approach seeks to fit a proxy functio
cs.LG updates on arXiv.org
HealthCAT: An Interpretable Encoder-only Transformer Framework for Health Indicator Prediction and Temporal Interpretation of Wearable Sensor Data
・arXiv:2607.27635v1 Announce Type: cross Abstract: Wearable sensors continuously capture fine-grained multivariate time-series data, providing opportunities to model behavioural patterns associated with health outcomes.
・However, existing deep learning methods prioritise predictive accuracy over interpretability, limiting their application in health research.
・In this study, we present HealthCAT, a flexible framework th
Zennの「大規模言語モデル」のフィード
Hermesで学ぶAIエージェント
・本記事は、早稲田AI研究会の勉強会資料として作成したものです。
・勉強会向けに内容を絞っているため、詳細な説明は省略している部分があります。気になったトピックがあれば、ぜひ関連記事や論文も参照してみてください。
・Hermesで学ぶAIエージェント入門 最近、AIエージェントが話題になっていますよね。ただ、ChatGPTのような普通のチャット型AIと何が違うのか、ピンとこない人も多いと思います。
cs.LG updates on arXiv.org
Heterogeneous Ranking in Industrial-Scale Recommender Systems: A Case Study
・arXiv:2607.27577v1 Announce Type: cross Abstract: Heterogeneous recommendation feeds present complex challenges that extend beyond those found in highly homogeneous environments (e.g., music-only or video-only closed-ecosystem platforms).
・In Google Discover, a unified feed integrates diverse content sourced from the decentralized open web, including web articles, long-form and short-form videos, user-generated conten
stat.ML updates on arXiv.org
Heterogeneous Treatment Effect Estimation under Noncompliance in the Illinois Workplace Wellness Study with Bayesian Tree Ensembles
・arXiv:2408.07765v3 Announce Type: replace-cross Abstract: Estimating varying treatment effects in randomized trials with noncompliance is inherently challenging since variation comes from two separate sources: variation in the impact itself and variation in the compliance rate.
・In this setting, existing flexible machine learning methods are sensitive to the weak instruments problem and can yield unstable estimates of
cs.LG updates on arXiv.org
Hierarchical Multilevel Monte Carlo for Order-Optimal Neural Actor-Critic in Average-Reward CMDPs
・arXiv:2607.28390v1 Announce Type: new Abstract: Constrained Markov Decision Processes (CMDPs) provide a natural framework for reinforcement learning in safety-critical applications, where agents maximize long-term reward while satisfying long-term constraints.
・Although primal-dual actor-critic methods with linear critics are well understood, extending order-optimal convergence guarantees to neural critics in average-
cs.LG updates on arXiv.org
HOMER: Huber-of-Means for Efficient and Robust Estimation in Hilbert Spaces
・arXiv:2607.27532v1 Announce Type: cross Abstract: Heavy tails weaken high-confidence control for the empirical mean.
・Geometric median-of-means (MOM) also lacks a threshold that moves toward mean efficiency.
・We propose \emph{HOMER}, or Huber-of-Means for Efficient and Robust Estimation.
OpenAI News
How avatarin built a 24/7 retail agent with GPT-Realtime
・avatarin uses OpenAI’s GPT-Realtime to give Yamada Denki shoppers 24/7 multilingual support.
・In two weeks, 30,000 people used the agent and 92% of survey responses were positive.
cs.LG updates on arXiv.org
How Can We Synthesize High-Quality Pretraining Data? A Systematic Study of Prompt Design, Generator Model, and Source Data
・arXiv:2604.13977v2 Announce Type: replace-cross Abstract: Synthetic data is a standard component in training large language models, yet systematic comparisons across design dimensions, including rephrasing strategy, generator model, and source data, remain absent.
・We conduct extensive controlled experiments, generating over one trillion tokens, to identify critical factors in rephrasing web text into synthetic pretra
cs.LG updates on arXiv.org
Improved Classification of Nitrogen Stress Severity in Plants Under Combined Stress Conditions Using Spatio-Temporal Deep Learning Framework
・arXiv:2509.06625v3 Announce Type: replace-cross Abstract: Plants in their natural habitats endure an array of interacting stresses, both biotic and abiotic, that rarely occur in isolation.
・Nutrient stress-particularly nitrogen deficiency-becomes even more critical when compounded with drought and weed competition, making it increasingly difficult to distinguish and address its effects.
・Early detection of nitrogen str
cs.LG updates on arXiv.org
Improving the Robustness/Accuracy Tradeoff Against Adversarial Attacks Using Information Bottleneck Distillation Through Dual Teachers
・arXiv:2607.27737v1 Announce Type: new Abstract: Deep neural networks (DNNs) have achieved remarkable success in classical machine learning problems.
・However, they are known to be vulnerable to adversarial attacks.
・Countermeasures proposed in the literature, notably Information Bottleneck Distillation (IBD) introduced by Kuang et al., degrade the classification accuracy on clean inputs while improving the robustness t
Takara TLDR - Daily AI Papers
Inducing language models to assert their own consciousness restores human beliefs and values
・Aligning large language models to prevent them attributing consciousness to themselves inadvertently alters their representations of mindedness in other entities alongside human beliefs and values.
・We demonstrate that safety fine-tuning suppresses models' tendencies to attribute minds not only to themselves, but also to non-human animals and natural objects, while also driving a reduction in spiritual belief.
Takara TLDR - Daily AI Papers
InfoOps Bench: A live information operations safety benchmark
・In this paper we present an active, constantly updated AI benchmark which measures the integrity of frontier language models against being co-opted for state-backed information operations.
・We draw on over 2,100 information operations from a live monitoring pipeline which tracks Russian, Chinese and Iranian state-backed information assets.
・Alongside this paper, we release a companion website that tracks the most promi
cs.LG updates on arXiv.org
Information Bottleneck Learning for Faithful Time Series Forecasting Explanations
・arXiv:2607.28124v1 Announce Type: new Abstract: As forecasts increasingly drive decisions in fields such as energy, transportation, and healthcare, understanding the historical data behind these predictions has become as crucial as the predictions themselves.
・Although existing interpretable-by-design forecasters reveal their internal structures, they offer no guarantee that these structures faithfully reflect the und
cs.LG updates on arXiv.org
Integrating Contextual Embeddings into Evaluation of Expressive MIDI Piano Performances
・arXiv:2607.27909v1 Announce Type: cross Abstract: Objective evaluation of expressive MIDI piano performances typically relies on attribute statistics such as timing, velocity, and duration of individual notes.
・However, these methods often disregard dependencies between notes, which poses a potential limitation in assessing the similarity between two sets of performances.
・In generative applications, the wide variety o
cs.LG updates on arXiv.org
Inverse design of bespoke interatomic potentials via active learning by information-matching
・arXiv:2606.08148v2 Announce Type: replace-cross Abstract: Interatomic potentials (IPs) enable large-scale atomistic simulations beyond the reach of first-principles methods, but their predictive reliability depends critically on the selection of training data, quantified uncertainty, and model expressiveness.
・Active learning (AL) provides a principled framework for constructing efficient and accurate IPs, yet most st
cs.LG updates on arXiv.org
It's All Just Vectorization: einx, a Universal Notation for Tensor Operations
・arXiv:2607.27987v1 Announce Type: new Abstract: Tensor operations represent a cornerstone of modern scientific computing.
・However, the Numpy-like notation adopted by predominant tensor frameworks is often difficult to read and write and prone to so-called shape errors, i.a., due to following inconsistent rules across a large, complex collection of operations.
・Alternatives like einsum and einops have gained popularity
＠IT 全フォーラム 最新記事一覧
ITインフラ専門家の私でも、もうAIには勝てない――「Active Directory」障害対応をAIに丸ごと任せてみた結果
・AIにメールや資料を書かせるのではなく、システムを変更できる管理者権限を渡し、ITインフラ障害の調査から対処まで任せたら、何が起きるのでしょうか。本連載「Microsoft MVP胡田のAI検証ラボ――情シス業務、AIにどこまで任せられるか」の第1回では、AIにActive Directoryの障害対応を丸ごと任せました。原因も調査手順も教えません。AIはどこまで自力で原因にたどり着き、どのようにシステムを変更したのか。実際の操作記録とともに検証します。
cs.LG updates on arXiv.org
KAISEN: Reproducible Subgroup Fairness Auditing for Clinical Risk Models
・arXiv:2607.28608v1 Announce Type: new Abstract: Clinical risk models routinely achieve strong aggregate performance while producing materially different error rates across patient subgroups.
・Audit pipelines have been proposed to catch this, but their components are rarely stress-tested, so it is unclear which parts of an audit can be trusted and under what conditions.
・We present KAISEN, a five-phase audit pipeline co
cs.LG updates on arXiv.org
Kalman Meets Curriculum: Efficient Dynamic Prompt Selection for Adaptive RL Finetuning
・arXiv:2607.27610v1 Announce Type: new Abstract: Reinforcement learning (RL) finetuning significantly enhances the reasoning capabilities of large language models (LLMs), yet its effectiveness critically depends on selecting prompts of appropriate difficulty for the current policy.
・This is challenging because prompt difficulty evolves throughout training.
・Existing online methods therefore face a trade-off: evaluation-
cs.LG updates on arXiv.org
KernelGenBench: A Multi-Source and Multi-Chip Benchmark for LLM-based Kernel Generation
・arXiv:2607.27231v1 Announce Type: cross Abstract: Large language models (LLMs) have significantly increased the demand for efficient accelerator kernels, but kernel development remains a highly specialized and labor-intensive task.
・The recent rise of LLMs and agentic frameworks offers a promising pathway toward automatic kernel generation.
・However, despite rapid progress, there is still no comprehensive benchmark to
cs.LG updates on arXiv.org
Kohn-Sham Spectral Embedding on Sparse Graphs at the Nishimori Temperature for Image Classification
・arXiv:2607.28428v1 Announce Type: new Abstract: We introduce Kohn--Sham Spectral Embedding (KSSE), a physics-inspired energy-based model replacing dense CNN classifiers with a sparse-graph spectral embedding evaluated at the Nishimori temperature of an associated Random-Bond Ising Model.
・By mapping pre-trained features onto quasi-cyclic low-density parity-check graphs and constructing a regularized Laplacian acting a
Takara TLDR - Daily AI Papers
Large scale cross-regional remote sensing flood monitoring framework for operative mapping and impact analysis
・Effective flood monitoring is critical for minimizing the impacts of flood disasters on populations and infrastructure.
・Yet reliable remote sensing across extensive and environmentally diverse regions remains challenging, as most segmentation algorithms lack the generalisation capacity required for large-scale application, while annotated flood data are scarce and unevenly distributed.
・This study presents an end-to-e
Takara TLDR - Daily AI Papers
LAST: The Last Query Token Guides Visual Token Pruning for Edge-Cloud Collaborative MLLM Inference
・Multimodal foundation models are reshaping edge-cloud visual intelligence from task-specific feature pipelines into token-based interfaces, where edge devices encode visual inputs into tokens for a general-purpose cloud MLLM.
・However, dense visual-token sequences increase cloud-side inference costs.
・Existing pruning methods mainly target centralized inference: vision-driven methods can operate before cloud execution
cs.LG updates on arXiv.org
Latent Matters: Learning Deep State-Space Models
・arXiv:2602.23050v2 Announce Type: replace Abstract: Deep state-space models (DSSMs) enable temporal predictions by learning the underlying dynamics of observed sequence data.
・They are often trained by maximising the evidence lower bound.
・However, as we show, this does not ensure the model actually learns the underlying dynamics.
cs.LG updates on arXiv.org
Latent States in Neural Networks: Recovering the Temporal Structure of Drifting Data from Model Weights
・arXiv:2607.27482v1 Announce Type: new Abstract: A temporally drifting data stream may pass through discrete regimes rather than changing continuously.
・We ask whether such regimes are recoverable from the weights of models trained on the stream, using a hidden Markov model (HMM) fit to the chronologically ordered trajectory of those weights.
・We study this question in two domains known to drift over time: multimodal mi
cs.LG updates on arXiv.org
Latent-Kernel Discrete Flow Maps for Few-Step Generation
・arXiv:2607.27529v1 Announce Type: new Abstract: Discrete diffusion and flow-matching models denoise a sequence over many steps, but to keep each step cheap, they factorize the transition across positions and decide every token independently.
・This makes few-step generation challenging for text when the target couples two positions, such as a subject and a verb that must agree.
・An independent update commits to them sep
cs.LG updates on arXiv.org
Learning features from Newton's algorithm: a way to accelerate nonlinear parametrized PDE solvers
・arXiv:2607.28036v1 Announce Type: new Abstract: It is well known that Newton's method converges faster when the initial guess is closer to a root of a system of nonlinear equations.
・In this paper, a two-stage Newton initial guess strategy is proposed by learning features from a parameter-space sampling and a database of precomputed solutions.
・The method uses discrete Newton trajectories to construct two complementary
cs.LG updates on arXiv.org
Learning the Helmholtz equation operator with DeepONet for non-parametric 2D geometries
・arXiv:2605.00760v2 Announce Type: replace Abstract: This paper deals with solving the 2D Helmholtz equation on non-parametric domains, leveraging a physics-informed neural operator network, the DeepONet framework.
・We consider a 2D square domain with an inclusion of arbitrary boundary geometry at its center.
・It acts as a scatterer for an incoming harmonic wave.
cs.LG updates on arXiv.org
Learning to Detect Cyber Attacks: Neural Anomaly Detection for Cybersecurity with Theoretical Insights
・arXiv:2409.08521v2 Announce Type: replace-cross Abstract: In cybersecurity practice, new forms of cyberattacks continuously emerge, deliberately designed to evade defense systems that rely on previously observed behaviors.
・Motivated by this challenge, we propose a neural network-based method for anomaly detection that does not rely on (1) prior knowledge of anomaly distributions or (2) the availability of real anomal
cs.LG updates on arXiv.org
Learning to Select, Not Relearn: Hard-Routed Mixtures of Reasoning LoRAs
・arXiv:2606.31413v2 Announce Type: replace-cross Abstract: Composing independently trained LoRA adapters into a single large language model is useful for multi-domain adaptation, especially when the original training data cannot be shared.
・A common approach is to use MoE-style routing over LoRA experts, but for frozen pretrained adapters, soft weighted combinations can change the unit-scale additive update under which
cs.LG updates on arXiv.org
Learning to Trace Seiberg Dualities
・arXiv:2607.28628v1 Announce Type: cross Abstract: Dualities play an important role in establishing both microscopic and emergent phenomena in a wide range of physical systems.
・In practice, though, it can often be computationally challenging to establish when two systems are dual, even when all of the "rules of the game" are well-known.
・Said differently, when confronted with two systems, how can one efficiently establ
Takara TLDR - Daily AI Papers
Learning to Understand Body Language from Flight through Robust 3D Avatar Placing
・Perceiving human motion and intent at long range is a prerequisite for socially intelligent aerial robots, yet the data to learn it barely exists.
・We introduce Drones2BodyLanguage, a dataset grounding human motion in real UAV footage: avatars manifesting ten communicative intents are placed into unmodified 4K drone scenes with metrically correct position, scale and orientation, maintained over hundreds of frames of c
cs.LG updates on arXiv.org
Learning-Augmented Algorithms for Online Vertex Cover
・arXiv:2606.22831v2 Announce Type: replace-cross Abstract: This paper studies learning-augmented online weighted vertex cover with local advice and a tradeoff parameter $\lambda \in (0,1)$.
・We consider two graph settings: bipartite graphs and general graphs.
・In both settings, the online algorithm must maintain a feasible vertex cover under irrevocable decisions.
cs.LG updates on arXiv.org
Learning-Augmented and Randomized Algorithms for Line Aggregation with Delays
・arXiv:2607.27807v1 Announce Type: new Abstract: This paper studies learning-augmented and randomized online aggregation with delays on a line metric.
・We consider advice given as online suggested service lengths, and evaluate the algorithms in terms of robustness and consistency.
・For each $\lambda \in (0,1]$, we first propose a deterministic learning-augmented \textsc{Balance} algorithm that is $(4/\lambda+1/\lambda^2
cs.LG updates on arXiv.org
LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger
・arXiv:2607.28374v1 Announce Type: new Abstract: Multimodal agents for visual question answering increasingly operate as multi-step trajectories that interleave perception, retrieval, and reasoning, yet evaluation still largely reduces to final-answer accuracy.
・This aggregate signal cannot tell whether a correct answer was reached through grounded evidence, language priors, or accidental error cancellation.
Takara TLDR - Daily AI Papers
LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger
・Multimodal agents for visual question answering increasingly operate as multi-step trajectories that interleave perception, retrieval, and reasoning, yet evaluation still largely reduces to final-answer accuracy.
・This aggregate signal cannot tell whether a correct answer was reached through grounded evidence, language priors, or accidental error cancellation.
・We propose to treat a multimodal agent trajectory as a pro
Takara TLDR - Daily AI Papers
LEEPS: Latent-Guided Explore-Exploit Prompt Sampling for Efficient RLVR in Large Language Models
・Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models, but prompt groups with identical rollout rewards consume generation budget without effective learning signals.
・Pre-rollout prompt selection can reduce this waste by screening prompts before rollout generation.
・However, existing pre-rollout methods struggle to balance exploitation and exploration: repeat
cs.LG updates on arXiv.org
LightRot: A Light-Weighted Rotation Scheme and Architecture for Accurate Low-Bit Large Language Model Inference
・arXiv:2607.27704v1 Announce Type: cross Abstract: As large language models (LLMs) continue to demonstrate exceptional capabilities across various domains, the challenge of achieving energy-efficient and accurate inference becomes increasingly critical.
・This work presents LightRot, a lightweight rotation scheme and dedicated hardware accelerator designed for low-bit LLM inference.
・The proposed architecture integrates
cs.LG updates on arXiv.org
Linear Strategic Classification with Endogenous Improvements
・arXiv:2606.01198v2 Announce Type: replace Abstract: Strategic classification studies settings in which agents respond to a deployed classifier by modifying observable features at a cost.
・Classical models typically treat such responses as cosmetic: features may change, but true labels remain fixed.
・We study an improvement-aware variant in which strategic responses can induce genuine changes in outcome-relevant feature
cs.LG updates on arXiv.org
LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints
・arXiv:2410.06458v2 Announce Type: replace-cross Abstract: Instruction following is a key capability for LLMs.
・However, recent studies have shown that LLMs often struggle with instructions containing multiple constraints (e.g.
・a request to create a social media post "in a funny tone" with "no hashtag").
Zennの「大規模言語モデル」のフィード
LLM-as-a-Verifierが提起する検証ループの設計論
・検証（verification）を、事後学習・テスト時計算に続く新しいスケーリング軸として扱う。
・同じ問題を解いた5つのコード案があるとする。どれが一番正しいか、あなたはどう選ぶだろうか。
・多くのAIエージェントは「1〜10点で採点して」とLLMに聞き、一番高い点数の案を選ぶ。
cs.LG updates on arXiv.org
LLM-Guided Initialization for Accelerated Hybrid Quantum-Classical Medical Image Classification
・arXiv:2607.27262v1 Announce Type: cross Abstract: Variational quantum algorithms often encounter barren plateaus, where cost gradients decay rapidly with increasing circuit depth, undermining the trainability of parameterized quantum circuits.
・This paper evaluates AdaInit (Adaptive Initialization), proposed by Zhuang and Cunningham, which uses large language models to propose initial parameters for quantum neural net
Zennの「大規模言語モデル」のフィード
LLM-jp-Moshi-v1 を AWS EC2 と SSM ポートフォワードで安全に検証してみる
・はじめに 従来の音声対話システムでは、ユーザーの発話を受け取ってから応答を生成・読み上げる構成が多く、会話中の割り込みや相槌のような自然なやり取りが難しいという課題があります。そこで LLM-jp-Moshi-v1 を AWS の GPU EC2 上で起動し、手元のブラウザから full-duplex 音声対話 (送信と受信を同時に行える音声対話) を試してみました。
・本記事では、次の 2 点を中心に検証・整理しています。
・AWS のGPU EC2 上で LLM-jp-Moshi-v1 を動かし、ブラウザから音声対話できるかを確認する 「技術的に正しく動くか」と「実業務で使える品...
Zennの「大規模言語モデル」のフィード
LLMに非構造テキストをタグ化させたら平然と推測で埋めてきたので、コードで止めた
・個人開発で「愛犬と泊まれる宿を条件で絞り込むアプリ」を作りかけて、データを126件集めた時点でボツにしました。
・プロダクトとしては失敗ですが、途中で踏んだ地雷が全部再利用できる知見だったので技術面だけ切り出して残します。
・企画面の話は note のほう に書きました。この記事は実装の話だけです。
Zennの「大規模言語モデル」のフィード
LLMマルチエージェントの深層：連携とFunction Callingの設計原則
・LLMマルチエージェントの深層：連携とFunction Callingの設計原則 「LLMに複雑なタスクを任せたいけど、単一のプロンプトでは限界がある…」「外部ツールと連携させたいけど、どう設計すればいいかわからない…」そう感じていませんか？ 多くのエンジニアが直面するこの課題は、LLM単体では難しい複雑なタスクを、複数のエージェントが連携して解決するマルチエージェントシステムと、外部ツール連携の鍵となるFunction Calling（またはTool Calling）を効果的に活用することで克服できます。この記事では、これらの設計思想と実践的な実装方法を深掘りし、高度なAIアプリケ...
cs.LG updates on arXiv.org
LM-GRASP: Instance-Specific Language Models for Combinatorial Construction via Online Imitation Learning
・arXiv:2607.28135v1 Announce Type: new Abstract: Machine learning for combinatorial optimization typically relies on neural constructors trained via reinforcement learning on large offline datasets for a fixed problem class-incurring high pretraining costs and generalizing poorly outside the training distribution.
・We propose an alternative: a metaheuristic framework that reformulates the randomized constructive phase
Takara TLDR - Daily AI Papers
LoMeVQA: A Comprehensive Benchmark for Longitudinal Medical VQA
・In clinical practice, patients often undergo multiple imaging examinations over successive visits, yielding longitudinal data.
・Modeling such temporal information is crucial for reliable assessment of disease progression and treatment response.
・However, despite the rapid advancement of multimodal large language models (MLLMs), longitudinal medical visual reasoning remains largely underexplored.
cs.LG updates on arXiv.org
Looped Transformers with Source-Centered State Evolution
・arXiv:2607.27656v1 Announce Type: new Abstract: Looped Transformers create a useful train- and test-time compute axis by reusing the same Transformer block over recurrent depth, increasing effective depth at a fixed parameter count.
・However, that shared block must then govern an entire trajectory of varying hidden states over trained and extrapolated depths.
・Furthermore, in additive-injection looped Transformers, an
cs.LG updates on arXiv.org
LoRA Scaffolded Policy Optimization (LSPO): A Sampling-Time Low-Rank Scaffold for Recovering Reinforcement-Learning Gradient on Zero-Reward Cliff Prompts
・arXiv:2607.27787v1 Announce Type: new Abstract: Reinforcement learning from verifiable rewards (RLVR) for mathematical reasoning suffers from a structural blind spot: on "cliff" prompts-those on which every sampled rollout in a group fails-the group-normalized advantage is identically zero, so GRPO produces no gradient on precisely the prompts at the frontier of the model's capability.
・We introduce LoRA Scaffolded Po
cs.LG updates on arXiv.org
MatCreatioNN: Machine learning-guided computational discovery of photocatalysts for environmental applications
・arXiv:2607.27295v1 Announce Type: cross Abstract: The rational design of photocatalysts for environmental remediation and CO2 conversion remains limited by the high computational cost and sparse experimental data describing multi-parameter photocatalytic behavior.
・This work presents an integrated machine-learning framework that couples reinforcement learning-based metal-organic framework (MOF) generation with a multi
cs.LG updates on arXiv.org
MDL-GBG: A Non-parametric and Interpretable Granular-Ball Generation Method for Clustering
・arXiv:2605.08759v3 Announce Type: replace Abstract: Existing granular-ball generation methods are still mainly driven by handcrafted quality measures and heuristic splitting or stopping criteria, which may weaken the transparency of local generation decisions in clustering.
・To address this issue, this paper proposes Minimum Description Length based Granular-Ball Generation (MDL-GBG), a non-parametric and interpretabl
cs.LG updates on arXiv.org
Measuring Distortion in the Empty Regions of Dimensionality Reduction Scatterplots with the Gap Index
・arXiv:2607.28324v1 Announce Type: new Abstract: Quality metrics play a crucial role in the proper use of dimensionality reduction projections for visual analysis of high-dimensional data.
・They quantify the degree of distortion of a projection compared to the high-dimensional data and provide a reliable indication of how confident users can be in the structures they see in the resulting layouts.
・However, most popular
cs.LG updates on arXiv.org
Memory Efficient Tabular Foundation Models
・arXiv:2607.27546v1 Announce Type: new Abstract: Tabular Foundation Models, such as TabPFN, have received a large amount of recent attention due to their performance on in-context tabular machine learning tasks, which often exceeds classical baselines.
・However, practical deployment considerations of these models has received less attention.
・In this paper we investigate the memory requirements for these models.
Takara TLDR - Daily AI Papers
Metaphor Tracer: A Theory-Informed Analysis of Hidden States
・What do a language model's hidden states say about the organization of a single text?
・From one forward pass, without training, we score every token position on two properties.
・The *aggregator* measures whether the position consolidates the whole text into a stable configuration.
cs.LG updates on arXiv.org
Metareasoning constraints couple narratives, affect and cognition
・arXiv:2502.09487v4 Announce Type: replace-cross Abstract: Narratives and emotions shape thoughts, and thoughts shape our feelings and stories we tell.
・Why narrative, affective and cognitive states interact remains unclear.
・We examine whether this mutual relationship reflects constraints on metareasoning - deciding what to think about - imposed by a shared computational state.
cs.LG updates on arXiv.org
Meteosat Third Generation imagery improves CNN-based SSI retrieval
・arXiv:2607.28093v1 Announce Type: cross Abstract: Accurate Surface Solar Irradiance (SSI) estimation is increasingly important for photovoltaic energy monitoring and forecasting.
・The recently introduced Meteosat Third Generation (MTG) satellite constellation provides imaging data with higher spatial resolution compared to the Meteosat Second Generation (MSG) satellite constellation, but its benefits for machine-learn
cs.LG updates on arXiv.org
MixFrag: Fragility-Guided Mixed-Precision Post-Training Quantization for Vision Transformers
・arXiv:2607.28589v1 Announce Type: cross Abstract: Post-training quantization (PTQ) has emerged as an effective solution for deploying Vision Transformers (ViTs) on resource-constrained devices.
・However, existing PTQ methods typically employ uniform bit-widths across transformer components, overlooking their heterogeneous sensitivity to quantization and leading to inefficient precision allocation.
・In this paper, we pr
cs.LG updates on arXiv.org
Modeling Decisions in Blockchain Analytics: A Leakage-Aware Evaluation of Tree-Based vs. Sequential Models
・arXiv:2607.27350v1 Announce Type: new Abstract: Sybil bots are Ethereum actors that imitate legitimate users to extract airdrop rewards or influence governance.
・Recent Sybil detection methods increasingly use deep learning and treat blockchain activity as a quasi-linguistic sequence.
・However, complex sequence models are computationally expensive for real-time monitoring, and their reported performance may be inflated
Takara TLDR - Daily AI Papers
MonoVoc: Decoupling Geometry and Semantics for Lightweight Monocular Open-Vocabulary 3D Gaussians
・Open vocabulary 3D scene understanding is essential for next-generation interactive systems, empowering users to intuitively query and navigate reconstructed environments using natural language.
・However, current 3D Gaussian frameworks are often bottlenecked by restrictive multiview capture requirements, costly scene-specific optimization, and the massive memory overhead of storing dense language features.
cs.LG updates on arXiv.org
MOON2.0: Dynamic Modality-balanced Multimodal Representation Learning for E-commerce Product Understanding
・arXiv:2511.12449v3 Announce Type: replace-cross Abstract: Recent Multimodal Large Language Models (MLLMs) have significantly advanced e-commerce product understanding.
・However, they still face three challenges: (i) the modality imbalance induced by modality mixed training; (ii) underutilization of the intrinsic alignment relationships among visual and textual information within a product; and (iii) limited handling o
cs.LG updates on arXiv.org
More Data, Worse Decisions? Preference Reversals in Neural Networks under Gram Incompatibility
・arXiv:2607.27255v1 Announce Type: cross Abstract: Neural networks increasingly combine data across populations, time periods, and operating conditions to improve generalization.
・This raises a reliability question: whether a model refitted on pooled data preserves an action ordering supported by both sources.
・Case-Based Decision Theory (CBDT) formalizes this requirement through its composition axiom, which requires so
cs.LG updates on arXiv.org
MORFES: A Benchmark for Productive Inflectional Competence in Modern Greek
・arXiv:2607.28274v1 Announce Type: cross Abstract: Modern Greek is a richly inflected language, yet the language models built for it are evaluated mainly on factual knowledge, and no benchmark is dedicated to their inflectional competence.
・We introduce MORFES (Morphological Open-class Recognition-and-Formation Evaluation Suite), a benchmark of 500 expert-verified items that tests the recognition and production of Gree
cs.LG updates on arXiv.org
MSGNN: A Spectral Graph Neural Network Based on a Novel Magnetic Signed Laplacian
・arXiv:2209.00546v5 Announce Type: replace-cross Abstract: Signed and directed networks are ubiquitous in real-world applications.
・However, there has been relatively little work proposing spectral graph neural networks (GNNs) for such networks.
・Here we introduce a signed directed Laplacian matrix, which we call the magnetic signed Laplacian, as a natural generalization of both the signed Laplacian on signed graphs and
cs.LG updates on arXiv.org
MUGEN: A Unified Framework for Efficient Motion Understanding and Generation
・arXiv:2607.27581v1 Announce Type: new Abstract: Grounding human motion in language, and language in motion, is a central step toward physical AI systems that can understand, generate, and communicate human behavior.
・Unified motion--language systems first coupled the two directions through a shared discrete motion codebook, but quantization limits generation quality.
・The strongest generators buy quality back at growin
cs.LG updates on arXiv.org
Multi-channel Uplift Policy Learning
・arXiv:2607.28182v1 Announce Type: new Abstract: E-commerce platforms must allocate fixed marketing budgets across multiple channels to maximize business utility.
・However, standard predict-then-optimize (PTO) paradigms fail in this compositional space due to observational confounding and severe extrapolation.
・We formulate this challenge as a simplex-constrained uplift decision problem and propose ReAlloc, a fast-slow
cs.LG updates on arXiv.org
Nanoparticle Networks for Neuromorphic Computing
・arXiv:2607.27844v1 Announce Type: cross Abstract: Physical computing leverages complex dynamical systems for energy-efficient data processing.
・In this work, we present a neuromorphic architecture based on metallic nanoparticles interconnected by molecular junctions on a $\text{SiO}_2$/Si substrate.
・We demonstrate that surrounding static control electrodes transform this nanoparticle network from a passive reservoir i
cs.LG updates on arXiv.org
Negative controls reveal volume-driven confounding in radiomics and imaging foundation model features
・arXiv:2607.28423v1 Announce Type: cross Abstract: Radiomics and imaging foundation models promise non-invasive biomarkers of tumour biology, yet predictive signatures may reflect tumour volume or acquisition artifacts rather than meaningful image structure.
・We introduce READII-2-ROQC, an open-source framework that uses volume-preserving negative controls to assess whether radiomic and deep imaging features capture in
cs.LG updates on arXiv.org
Neural Network Approximation of Solutions to Fractional Parabolic Partial Differential Equations
・arXiv:2607.27781v1 Announce Type: cross Abstract: We establish a dimension-efficient neural network approximation theory for solutions to fractional parabolic equations with lower-order drift and potential terms.
・By introducing anisotropic spectral Barron spaces, which measure temporal and spatial regularity separately in frequency space, we first develop a dimension-independent maximal regularity theory for these eq
cs.LG updates on arXiv.org
Neural Network-Assisted CLEAN for Channel Modeling in Low-SNR Regimes
・arXiv:2607.27450v1 Announce Type: new Abstract: Accurate multipath parameter estimation is critical for modern wireless communication systems, particularly in challenging low-SNR environments.
・Traditional Maximum Likelihood Estimation algorithms, such as CLEAN, provide high-resolution parameter extraction but suffer from prohibitive computational complexity due to exhaustive grid search.
・Conversely, purely data-drive
cs.LG updates on arXiv.org
Neurosymbolic Imitation Learning with Human Guidance: A Privileged Information Approach
・arXiv:2605.07166v2 Announce Type: replace Abstract: Imitation learning is widely used for learning to act in complex environments.
・While pure neural-based methods handle high dimensional data effectively, they suffer from the requirement of large number of samples and are prone to overfitting.
・Pure symbolic approaches, while generalize well, do not handle high-dimensional data effectively.
cs.LG updates on arXiv.org
NMINE: Normalized Mutual Information Neural Estimation
・arXiv:2607.27710v1 Announce Type: new Abstract: Mutual information is a general measure of statistical dependence that captures both linear and nonlinear relationships between random variables.
・For continuous and multidimensional variables For continuous multidimensional variables, mutual information must be estimated from samples.
・Because mutual information is unbounded, its values are not directly comparable across
cs.LG updates on arXiv.org
Noisy Data is Destructive to Reinforcement Learning with Verifiable Rewards
・arXiv:2603.16140v2 Announce Type: replace Abstract: Reinforcement learning with verifiable rewards (RLVR) has driven recent capability advances of large language models across various domains.
・Recent studies suggest that improved RLVR algorithms allow models to learn effectively from incorrect annotations, achieving performance comparable to learning from clean data.
・In this work, we show that these findings are inva
stat.ML updates on arXiv.org
Non-partitioned e-detectors for nonparametric sequential change detection
・arXiv:2607.28322v1 Announce Type: cross Abstract: We study the problem of sequential change detection over a general class of probability distributions ($\mathcal P$), where both the pre-change and post-change distributions are unknown and belong to $\mathcal P$.
・We do not assume a pre-specified partition of $\mathcal P$ into pre- and post-change families.
・We propose a general class of sequential change detectors obt
Takara TLDR - Daily AI Papers
Objective-Aligned Direct Answer SFT for Robust Multi-Frame Medical VQA
・Multi-frame medical VQA appears to reward increasingly complex adaptation: controller-style inference, localization-aware reranking, static hard-negative mixing, and staged continuation all appear plausible from first principles.
・We test a simpler competing hypothesis on MedFrameQA: methods that remain tightly aligned with the benchmark's final answer objective should be the strongest \emph{robust} adaptation family
cs.LG updates on arXiv.org
ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow
・arXiv:2607.27924v1 Announce Type: new Abstract: In the physical world we inhabit, space and time are fundamentally continuous.
・However, existing machine learning paradigms for world modeling are largely confined to discrete-time prediction, thereby exhibiting significant inefficiency in capturing the dynamics of physical world.
・We introduce Physical-Time Flow (\textbf{PT-Flow}), a novel approach that learns a continu
MachineLearningMastery.com
Ollama vs. LM Studio vs. llama.cpp: Which Local AI Runtime Should You Use in 2026?
・In this article, you will learn how Ollama, LM Studio, and llama.cpp differ across the dimensions that matter most to practitioners, and how to choose...
cs.LG updates on arXiv.org
On a joint simultaneous learning of relevant feature subsets and subspaces in regression-like problems
・arXiv:2607.28080v1 Announce Type: cross Abstract: We extend a recently introduced Entropy-Optimal Manifold Clustering (EOMC) to allow for a joint simultaneous identification of subsets and subspaces of relevant features in nonstationary and nonlinear regression problems.
・It is shown that the proposed extension - that we coin as Entropy-Optimal Manifold Regression (EOMR) - allows a robust learning with linearly-scalin
cs.LG updates on arXiv.org
On the Rate of Convergence of Kolmogorov-Arnold Network Regression Estimators
・arXiv:2509.19830v3 Announce Type: replace Abstract: Kolmogorov-Arnold Networks (KANs) approximate multivariate functions by composing univariate transformations through additive or multiplicative aggregation.
・We establish convergence guarantees for KANs whose univariate components are B-splines.
・The least-squares estimator over the KAN spline sieve attains the rate $O((\log n / n)^{2r/(2r+1)})$, uniformly over a ball
cs.LG updates on arXiv.org
On-Policy and Off-Policy Learning for Large Action Spaces
・arXiv:2607.28408v1 Announce Type: new Abstract: This thesis studies policy learning in interactive systems where an agent observes a context, selects an action from a very large set, and receives partial feedback.
・The main framework is contextual bandits, with two paradigms: on-policy learning, where the agent interacts sequentially with the environment and minimizes regret, and off-policy learning, where it learns f
Takara TLDR - Daily AI Papers
One Human, $N$ Agents: Audit-Budget Allocation for LLM Agent Fleets under Miscalibrated, Correlated Confidence
・A single human must audit $N$ LLM agents under a budget of $B \ll N$ audits per round, guided by self-reported confidence that may be adversarially miscalibrated and by correlated errors.
・We model this as budgeted noisy inspection over a two-level Gaussian copula and locate the miscalibration threshold $δ^*$ past which confidence-ranked auditing is \emph{worse} than random.
・Two a-priori expectations reverse: $δ^*$ \e
Takara TLDR - Daily AI Papers
One Patch Is Enough: Reinforcement-Optimized Visual Token Grounding for MLLM-Based Scene Text Spotting
・Scene text spotting requires high-precision alignment between textual recognition and spatial localization.
・While visual-token grounding has emerged as a promising formulation for Multimodal Large Language Models (MLLMs), the previous multi-patch paradigm often introduces redundant noise and localization ambiguity, particularly for dense or small text instances.
・To address this, we propose Single-Patch Text Spotting
cs.LG updates on arXiv.org
OneShot: Index-in-Ranking with Neural Scoring for Large-Scale Retrieval
・arXiv:2607.27475v1 Announce Type: cross Abstract: In modern recommendation systems, retrieval serves as a primary stage responsible for filtering billions of candidate items down to thousands prior to refined ranking.
・To make this massive search effective and efficient, the system relies on ranking accuracy and indexing efficiency.
・However, these two objectives are traditionally misaligned: while the former optimizes
cs.LG updates on arXiv.org
Open-Vocabulary BEV Segmentation with 3D-Aware Geometric Constraints
・arXiv:2606.24353v2 Announce Type: replace-cross Abstract: Bird's-eye view (BEV) perception fuses multi-camera images into a unified top-down representation for autonomous driving.
・Despite recent progress, state-of-the-art methods remain confined to closed-set scenarios, making them vulnerable to unpredictable real-world environments.
・In this work, we introduce open-vocabulary BEV segmentation (OVBS), which leverages
cs.LG updates on arXiv.org
Oracle-Budgeted Molecular Optimization with Short-Term Graph Memory
・arXiv:2607.28437v1 Announce Type: new Abstract: Molecular optimization is commonly performed under a limited oracle budget, which makes deciding what to evaluate as important as deciding what to generate.
・We introduce short-term graph memory, a plug-in module that preserves the generator architecture and native update rule while learning from previously evaluated molecules to prioritize subsequent oracle queries.
Takara TLDR - Daily AI Papers
Orca: Neural Operators for Causal Reasoning in Continuous Time
・Structural causal models are the standard language for reasoning about interventions and counterfactuals, but they describe static variables, typically measured once, and usually forbid cyclic dependencies.
・Many systems we care about, such as patients, climates, and economies, instead evolve continuously in time, are observed at irregular time points, and contain feedback loops.
・We argue that neural operator learning
cs.LG updates on arXiv.org
Persistent Gaussian Perturbations Prevent Oversmoothing in Recurrent Graph Neural Networks
・arXiv:2607.28185v1 Announce Type: new Abstract: Oversmoothing is a fundamental limitation of deep graph neural networks (GNNs), where repeated message passing causes node representations to become increasingly similar, eventually collapsing toward a low-dimensional subspace.
・This phenomenon limits the effective depth of message-passing architectures and motivates the search for mechanisms that preserve representation
Takara TLDR - Daily AI Papers
Persistent Gaussian Perturbations Prevent Oversmoothing in Recurrent Graph Neural Networks
・Oversmoothing is a fundamental limitation of deep graph neural networks (GNNs), where repeated message passing causes node representations to become increasingly similar, eventually collapsing toward a low-dimensional subspace.
・This phenomenon limits the effective depth of message-passing architectures and motivates the search for mechanisms that preserve representation diversity.
・In this paper, we study a recurrent
cs.LG updates on arXiv.org
Personalized RewardBench: Evaluating Reward Models with Human Aligned Personalization
・arXiv:2604.07343v2 Announce Type: replace-cross Abstract: Pluralistic alignment has emerged as a critical frontier in the development of Large Language Models (LLMs), with reward models (RMs) serving as a central mechanism for capturing diverse human values.
・While benchmarks for general response quality are prevalent, evaluating how well reward models account for individual user preferences remains an open challenge.
Takara TLDR - Daily AI Papers
PhiZero: A World Model Built Around Physical Language
・We introduce PhiZero, a physical world model built around physical language, a compact discrete representation of world-state transitions.
・Existing physical world models typically predict future videos directly in pixel space, leaving the underlying world dynamics implicit within high-dimensional visual predictors.
・Motivated by humans' ability to abstract predictive structure from visual experience and organize it in
cs.LG updates on arXiv.org
PlantBGC: Transformer for Plant BGC Discovery via Label-Free Domain Adaptation and Weak Supervision
・arXiv:2607.27258v1 Announce Type: cross Abstract: Plant biosynthetic gene clusters (BGCs) encode specialized-metabolite pathways, yet curated plant BGC labels remain scarce, hindering supervised discovery at genome scale.
・Existing plant BGC mining tools are largely signature- and rule-driven and do not fully leverage recent advances in contextual representation learning for modeling long-range domain context and cont
cs.LG updates on arXiv.org
PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective
・arXiv:2607.27265v1 Announce Type: new Abstract: Real-time bidding is central to computational advertising, comprising three elements: Supply Side Platform (SSP) selling ad impressions, Demand Side Platform (DSP) bidding for advertisers, and Ad Exchange conducting auctions between them.
・Traditional auto-bidding algorithms focus solely on the DSP side, maximizing advertiser conversions by adjusting bids against competi
cs.LG updates on arXiv.org
Policy Gradient Steering: Interventions from Behavioral Objectives
・arXiv:2607.27574v1 Announce Type: new Abstract: Activation steering has emerged in large language models as a lightweight alternative for dynamically changing a model's behavior at inference time.
・However, we show that existing steering methods fail to steer even a simple policy in a two-route gridworld environment.
・To address this limitation, we propose Policy Gradient Steering (PGS), which formulates steering as a
cs.LG updates on arXiv.org
Position, Not Provenance: Separating Reasoning Mediation from Sycophancy in Medical Vision-Language Models
・arXiv:2607.27304v1 Announce Type: new Abstract: Medical vision-language models (VLMs) generate chain-of-thought (CoT) reasoning before answering clinical questions, but whether this reasoning causally influences predictions remains unclear.
・We present CoT-Mediate, a behavioral framework that perturbs a single clinically meaningful attribute within a model's own generated reasoning and measures whether the resulting p
cs.LG updates on arXiv.org
Procedural Fairness in Multi-Agent Bandits
・arXiv:2601.10600v2 Announce Type: replace-cross Abstract: In the context of multi-agent multi-armed bandits (MA-MAB), fairness is often reduced to outcomes: maximizing welfare, reducing inequality, or balancing utilities.
・However, evidence in psychology, economics, and Rawlsian theory suggests that fairness is also about process and who gets a say in the decisions being made.
・We introduce procedural fairness as equal
cs.LG updates on arXiv.org
Prox: Training-Free FFN Activation Sparsity via Approximate Intermediate-Channel Salience in LLMs
・arXiv:2607.27591v1 Announce Type: new Abstract: Feed-forward networks (FFNs) dominate memory traffic and computation in large language model (LLM) inference, making them a primary target for activation sparsification.
・However, existing training-free methods suffer substantial model-quality degradation at high sparsity due to limitations in their channel-selection strategies.
・We observe that the SwiGLU intermediate st
cs.LG updates on arXiv.org
Psych-ECA: A Reproducible Semi-Synthetic Benchmark for Synthetic Control Arms in Longitudinal Psychiatry
・arXiv:2607.27224v1 Announce Type: cross Abstract: External and synthetic control arms (ECAs) are entering psychiatric drug development, but the field lacks a benchmark that evaluates the properties regulators care about: not only how accurately a method reconstructs untreated trajectories, but whether its uncertainty is calibrated, whether it is robust to the informative observation times common in mental-health reco
cs.LG updates on arXiv.org
QAdapt: A Noise-Adaptive Neural Pre-Decoding Framework for Quantum Error Correction
・arXiv:2607.28422v1 Announce Type: new Abstract: Fault-tolerant quantum computing (FTQC) relies on quantum error correction to suppress physical errors and preserve logical information at scale.
・In practice, however, performance is constrained not only by physical noise but also by the latency of classical decoders processing rapidly generated syndrome data.
・This challenge is exacerbated by hardware noise that is stro
cs.LG updates on arXiv.org
QQWorld: Quantile-Quantile Matching for World Model Regularization
・arXiv:2607.28415v1 Announce Type: new Abstract: Latent world models enable efficient planning by predicting future states in a compact representation space, but their performance depends critically on the quality of the learned latent distribution.
・LeWorldModel (LeWM) regularizes its latents toward an isotropic Gaussian using the Epps-Pulley (EP) objective.
・We show that the corrective gradients of EP rapidly vanish f
cs.LG updates on arXiv.org
Quadratic Objective Perturbation: Curvature-Based Differential Privacy
・arXiv:2605.05905v2 Announce Type: replace Abstract: Objective perturbation is a standard mechanism in differentially private empirical risk minimization.
・In particular, Linear Objective Perturbation (LOP) enforces privacy by adding a random linear term, while strong convexity and stability are ensured by an additional deterministic quadratic term.
・However, this approach requires the strong assumption of bounded gradi
cs.LG updates on arXiv.org
Quantum Speedups for Stochastic Optimization with Heavy-Tailed Noise
・arXiv:2607.25492v2 Announce Type: replace Abstract: We study stochastic optimization with heavy-tailed gradient noise.
・We first propose a novel quantum mean estimator for multivariate heavy-tailed random variables that achieves lower query complexity than optimal classical estimators in the low-dimensional regime.
・We further develop an unbiased quantum mean estimator by applying a generalized multi-level Monte Carlo
cs.LG updates on arXiv.org
QuantWAMs: Calibrating at the Right Granularity for World Action Models
・arXiv:2607.28405v1 Announce Type: cross Abstract: World Action Models (WAMs) jointly predict future observations and actions, but their iterative denoising and closed-loop execution make efficient deployment costly.
・Existing post-training quantization (PTQ) methods are poorly suited to WAMs because they rely on open-loop objectives, homogeneous model assumptions, and calibration distributions that do not reflect depl
cs.LG updates on arXiv.org
Random Projection Flows for Efficient Manifold Density Estimation
・arXiv:2509.25228v3 Announce Type: replace Abstract: Accurate density estimation is crucial for understanding complex high-dimensional data, but it becomes challenging when the data lies on or near low-dimensional manifolds.
・Random projections provide a natural way to reduce dimensionality while approximately preserving geometric structure, enabling effective density estimation in these settings.
・We introduce \emph{Ra
cs.LG updates on arXiv.org
Rao-Blackwellized Score Matching on Manifolds
・arXiv:2605.25567v3 Announce Type: replace-cross Abstract: We study denoising score matching (DSM) when data are drawn from an embedded manifold $M \subset \mathbb{R}^D$.
・We show that under ambient Gaussian corruption, the target has variance that diverges as the noise scale decreases and correct for it by regressing against the conditional expectation given the nearest point projection on the manifold: the $L^2$-opti
cs.LG updates on arXiv.org
Re-examining Granger Causality with Causal Bayesian Networks and Reichenbachs Principles
・arXiv:2501.02672v4 Announce Type: replace-cross Abstract: Granger causality (GC) is widely used to infer directed relationships in time-series data.
・However, its predictive criterion does not by itself distinguish direct causal effects from dependencies induced by common causes, indirect paths, collider conditioning, or model misspecification.
・We revisit this limitation by interpreting bivariate and multivariate GC t
cs.LG updates on arXiv.org
Reading Without a Reader: Large Language Models Collapse Reading and Writing into a Single Entangled Code
・arXiv:2607.24797v2 Announce Type: replace-cross Abstract: In the literate human brain, reading and writing doubly dissociate: a ventral decoding route (pure alexia) and a fronto-parietal encoding route (pure agraphia), sharing a partial orthographic core.
・A decoder-only large language model (LLM) drives both from one autoregressive path optimized on text (a \emph{cultural} invention, not an evolved instinct).
cs.LG updates on arXiv.org
Real-Time Hard Peak Age-of-Information Safety with No-Regret Learning
・arXiv:2607.27626v1 Announce Type: new Abstract: Safety-critical IoT systems such as industrial closed-loop control, V2X coordination, and remote teleoperation require every sensor's peak Age of Information (peak AoI, also abbreviated PAoI) to stay below a hard per-slot deadline, not merely an average bound.
・Existing approaches meet this requirement only under restrictive assumptions: stochastic channels for Whittle-i
cs.LG updates on arXiv.org
Reasoning Consensus: Structural Ensembling of LLM Reasoning via Weighted DAG Aggregation
・arXiv:2607.27783v1 Announce Type: cross Abstract: Large Language Models (LLMs) explore problems through chain-of-thought, but this exploration is buried in unstructured prose.
・On high-stakes tasks, users cannot tell which steps are well-supported, which alternatives were seriously considered, or how the final conclusion compares to those the model discarded.
・We propose a framework that ensembles the reasoning structu
cs.LG updates on arXiv.org
Recall Before You Rank: Similarity-Guided Top-$K$ Reuse for Efficient Long-Context Attention
・arXiv:2607.27692v1 Announce Type: cross Abstract: Top-$K$ sparse attention reduces the cost of Softmax and value aggregation by attending to only a small subset of key--value (KV) entries.
・However, identifying this subset still requires scoring the current query against the full KV cache and performing global Top-$K$ selection, leaving selector cost linear in context length and limiting the practical efficiency of sp
cs.LG updates on arXiv.org
Recognition and Label-Free Adaptation Across Recording Sessions in Surface-EMG Gesture Decoding
・arXiv:2607.27568v1 Announce Type: new Abstract: Recognition accuracy obtained during a recording session does not persist when a user puts on the electrodes again after the electrodes had previously been removed.
・The electrodes may have moved slightly, the skin may be drier or wetter, or the elbow may be positioned differently; these factors all contribute to day-to-day variability and therefore represent a major obs
cs.LG updates on arXiv.org
Recursive transformers for semiconductor thermo-mechanical reliability
・arXiv:2607.27251v1 Announce Type: new Abstract: Transformer-based surrogate models are increasingly used to replace expensive first-principles simulation in engineering design.
・But conventional transformer architectures are often over parameterized for the small, low-dimensional datasets typical of engineering design spaces, where large simulation data is expensive to generate.
・Under these conditions, excess paramete
cs.LG updates on arXiv.org
REFINE-DP: Diffusion Policy Fine-tuning for Humanoid Loco-manipulation via Reinforcement Learning
・arXiv:2603.13707v3 Announce Type: replace-cross Abstract: Humanoid loco-manipulation requires coordinated task-space motion planning with stable loco-manipulation command tracking under complex robot-environment dynamics and long-horizon tasks.
・While diffusion policies (DPs) show promise for learning from demonstrations, deploying them on humanoids poses critical challenges: the motion planner trained offline is deco
cs.LG updates on arXiv.org
Reflected diffusion, no-flux continuity equations and confined Lagrangian flows in bounded domains
・arXiv:2607.28344v1 Announce Type: cross Abstract: Motivated by marginal distribution flows of reflected diffusions in bounded domains, we investigate when a density/flux pair solving a no-flux continuity equation admits a regular Lagrangian flow that remains in the closed domain and generates the prescribed density flow.
・We give sufficient conditions in terms of interior bounded-variation regularity, bounded-variatio
cs.LG updates on arXiv.org
Region-adaptable retrieval of coastal biogeochemical parameters from near-surface hyperspectral remote sensing reflectance using physics-aware meta-learning
・arXiv:2605.05623v2 Announce Type: replace Abstract: Hyperspectral in situ sensing has shown promise in retrieving aquatic biogeochemical (BGC) parameters, such as total suspended solids, dissolved organic carbon, and total chlorophyll-a, for cost-effective monitoring of coastal water quality.
・However, generalising such retrieval algorithms across water bodies remains challenging, as the relationship between remote se
cs.LG updates on arXiv.org
Regularizing modality contribution drift in multimodal continual learning
・arXiv:2607.27260v1 Announce Type: new Abstract: Multimodal continual learning (MMCL) aims to learn emerging knowledge from multimodal data while preserving knowledge.
・To mitigate forgetting, current MMCL methods usually focus on cross-modal representation alignment or semantic similarity, but they overlook whether the relative contributions of individual modalities and their interactions remain stable across incremen
cs.LG updates on arXiv.org
Representation and Invariance in Reinforcement Learning
・arXiv:2112.07752v4 Announce Type: replace-cross Abstract: Researchers have formalized reinforcement learning (RL) in different ways.
・If an agent in one RL framework is to run within another RL framework's environments, the agent must first be converted, or mapped, into that other framework.
・In this paper, we lay foundations for studying relative-intelligence-preserving mappability between RL frameworks.
cs.LG updates on arXiv.org
Rethinking EEG-Based Disease Diagnosis: Decoupling Instance Representation Learning from Subject-Level Supervision
・arXiv:2607.27274v1 Announce Type: new Abstract: EEG-based disease diagnosis requires one prediction per subject, yet common pipelines segment recordings into short instances, inherit the subject label for every instance, and train instance-level classifiers.
・This assumes that all instances provide equally reliable diagnostic evidence.
・Multiple instance learning (MIL) avoids inherited labels by treating each subject a
Takara TLDR - Daily AI Papers
Rethinking LLM-Judged Helpfulness as a Pedagogy Signal: A Pre-Registered Audit Across Tutor Models
・LLM tutoring poses a measurement problem: can a general-purpose helpfulness rubric distinguish direct answer-giving from pedagogical guidance?
・We audit this signal in a pre-registered study.
・Within each of three tutor bases, we compare conversational and pedagogical policies instantiated with the same underlying model and paired with one fixed weak simulated student.
cs.LG updates on arXiv.org
ReToken: One Token to Improve Vision-Language Models for Visual Retrieval
・arXiv:2607.28627v1 Announce Type: cross Abstract: Long visual context poses a challenge for vision-language models: performance degrades as the number of distractors grows, and processing all tokens at once is computationally infeasible under GPU memory constraints.
・We present ReToken, a single learnable embedding trained as an explicit retrieval target that selects a sparse set of query-relevant visual tokens from a
cs.LG updates on arXiv.org
Reviewer Scores Are Not Comparable Across Research Areas in ML Peer Review
・arXiv:2607.27209v1 Announce Type: cross Abstract: Peer review at ML conferences increasingly relies on reviewer scores as the primary decision instrument.
・As submissions have scaled from thousands to tens of thousands per year, no systematic audit has examined whether this instrument functions uniformly across research areas, or whether acceptance outcomes are in practice shaped by forces that reviewer scores neither
cs.LG updates on arXiv.org
Revisiting Predictive Process Monitoring in the Age of Foundation Models: A Comparative Study of Sequence, Tabular, and LLM Approaches
・arXiv:2607.27797v1 Announce Type: new Abstract: Predictive process monitoring (PPM) leverages event logs to forecast the future of running process instances, for instance, predicting the next activity, the remaining time until case completion, or the time to the next event.
・While PPM research in recent years has been dominated by deep sequence models trained from scratch, such as Long Short-Term Memory (LSTM) models,
cs.LG updates on arXiv.org
RIPPLE: Generating Multi-Channel Phase, Not Recovering It
・arXiv:2607.27775v1 Announce Type: new Abstract: Generative models synthesize magnitude spectra with high fidelity, while phase is delegated to a recovery module---Griffin--Lim, a vocoder, or a latent decoder---applied independently to each channel.
・For multi-channel waveforms this delegation is costly: the physical content of spatial audio and three-component seismograms lives in the phase relationships between chann
cs.LG updates on arXiv.org
RLPF: Reinforcement Learning from Performance Feedback for Code Generation
・arXiv:2607.27271v1 Announce Type: new Abstract: Code models are increasingly trained with execution feedback, but most training signals still stop at correctness.
・This leaves an important gap for systems code: two programs can pass the same tests while differing greatly in runtime.
・We study how to train code agents to prefer faster correct implementations, rather than treating efficiency only as an evaluation metric.
cs.LG updates on arXiv.org
Robust Estimation of Sparse Numerical Vectors under Local Differential Privacy
・arXiv:2607.27815v1 Announce Type: cross Abstract: Local differential privacy (LDP) protocols are vulnerable to poisoning attacks.
・Existing research have proposed efficient defense strategies for single-item users.
・However, in practice, a user may possess multiple items.
cs.LG updates on arXiv.org
Robust Wavelength Selection for Partial Least Squares Sugar Content Estimation Using Combinatorial Bayesian Optimization
・arXiv:2607.27645v1 Announce Type: cross Abstract: Wavelength selection is one of the important preprocessing methods in near-infrared spectroscopy to improve prediction accuracy and interpretability of spectral data.
・We formulate wavelength-region selection for sugar content estimation as a binary black-box optimization problem and propose a method based on Bayesian optimization.
・The proposed method constructs a spar
cs.LG updates on arXiv.org
ROCS: Request-Oriented Compute Sharing for Efficient Large-Scale Recommendation
・arXiv:2607.27744v1 Announce Type: new Abstract: Modern recommendation models gain prediction quality by scaling feature-interaction and sequence modules, but production cost constraints cap how far systems can scale.
・In this work, we propose Request-Oriented Compute Sharing (ROCS), a modeling and inference paradigm that exploits a unique property of recommendation inference: each user request is evaluated against man
cs.LG updates on arXiv.org
S-CEReBrO: Breaking the Memory Barrier in Continuous EEG Monitoring
・arXiv:2607.27913v1 Announce Type: new Abstract: Foundation models offer a promising paradigm for Electroencephalography (EEG) analysis, leveraging generalizable representations from vast unlabeled datasets.
・Yet, Transformer-based architectures face a critical bottleneck: global attention mechanisms couple the attention memory state to the signal duration, causing memory overflow during continuous monitoring.
Takara TLDR - Daily AI Papers
S-CEReBrO: Breaking the Memory Barrier in Continuous EEG Monitoring
・Foundation models offer a promising paradigm for Electroencephalography (EEG) analysis, leveraging generalizable representations from vast unlabeled datasets.
・Yet, Transformer-based architectures face a critical bottleneck: global attention mechanisms couple the attention memory state to the signal duration, causing memory overflow during continuous monitoring.
・To address this, we introduce S-CEReBrO (Streaming CEReB
cs.LG updates on arXiv.org
S-GRPO: Unified Post-Training for Large Vision-Language Models
・arXiv:2604.16557v2 Announce Type: replace Abstract: Current post-training methodologies for adapting Large Vision-Language Models (LVLMs) generally fall into two paradigms: Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL).
・Despite their prevalence, both approaches suffer from inefficiencies when applied in isolation.
・SFT forces the model's generation along a single expert trajectory, often inducing catast
cs.LG updates on arXiv.org
Safety-Gated Agentic Supervisory Control on a Coupled Distillation Benchmark: Regime Map, Auditable Gate, and Co-Design Findings
・arXiv:2607.27849v1 Announce Type: cross Abstract: An open-weight LLM can write composition setpoints every five minutes.
・What a plant still needs is a hard check: named constraints, logged margins, and an admit/block decision before the regulatory layer moves.
・This paper puts that check in a rule-based forked-twin counterfactual gate (nine pinned constraints) and leaves the regulatory layer unchanged.
cs.LG updates on arXiv.org
Same Graph Cross-Task Transfer in GNNs: Protocols and Predictors
・arXiv:2607.28525v1 Announce Type: new Abstract: Many real-world graphs support multiple predictive tasks over the same underlying structure, creating an opportunity to reuse supervision across node classification (NC) and link prediction (LP).
・However, existing evaluations often rely on incompatible splits, observed-graph assumptions, and negative sampling rules, making conclusions about same-graph cross-task transfe
cs.LG updates on arXiv.org
Sample More, Reflect Less: Self-Refine and Reflexion Lose to Repeated Sampling at Equal Token Cost, from 1.5B to 7B
・arXiv:2607.28576v1 Announce Type: cross Abstract: Methods that make a language model plan, criticise and rewrite its own answer, reflect on mistakes, pick the best of several attempts, or debate with copies of itself nearly all make it generate far more text than a single chain of thought.
・Because generating more text raises accuracy by itself, a gain over one chain of thought does not show the method's idea is what
cs.LG updates on arXiv.org
ScaFE: Data-Efficient Scar Classification with LLM-Generated Clinical Feature Programs
・arXiv:2607.28538v1 Announce Type: cross Abstract: Classifying pathological scars from clinical photographs requires distinguishing keloids from hypertrophic scars despite limited expert-labeled data and substantial acquisition variation across hospitals.
・End-to-end image models remain data-dependent, whereas sending photographs to a hosted vision-language model (VLM) may conflict with local data-governance requiremen
cs.LG updates on arXiv.org
Scalable Drift Monitoring in Medical Imaging AI
・arXiv:2410.13174v3 Announce Type: replace-cross Abstract: The integration of artificial intelligence (AI) into medical imaging has advanced clinical diagnostics but poses challenges in managing model drift and ensuring long-term reliability.
・To address these challenges, we develop MMC+, an enhanced framework for scalable drift monitoring, building upon the CheXstray framework that introduced real-time drift detection
stat.ML updates on arXiv.org
Scalable Graph Coreset Selection via Greedy Sampling
・arXiv:2607.27602v1 Announce Type: cross Abstract: Sampling representative nodes from large graphs is fundamental to graph signal processing and network analysis, yet existing methods require access to the full graph Laplacian, making them impractical at scale.
・We propose a simple and effective column-selective graph sampling algorithm based on a minimum inner product greedy selection rule.
・At each iteration, the algo
cs.LG updates on arXiv.org
Schreier-Coset Graph Rewiring
・arXiv:2607.27479v1 Announce Type: new Abstract: The information flow in the graph neural networks (GNNs) is fundamentally constrained by over-squashing, where structural bottlenecks impede long range information propagation.
・Graph-rewiring methods, which modify graph topology, have been extensively used to alleviate this.
・However, existing approaches often introduce prohibitive structural and computational bottleneck
The latest research from Google
Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence
Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence
cs.LG updates on arXiv.org
SCOPE: Supply-Chain Operations through Coupled Policies for End-to-End Coordination
・arXiv:2607.28488v1 Announce Type: cross Abstract: Can supply-chain AI move beyond isolated decision modules toward unified operational planning?
・A complete replenishment plan specifies which products each location carries, which upstream facility supplies it, how often it is replenished, and how deliveries are routed.
・These decisions are operationally coupled: the selected assortment changes the demand and load passe
Takara TLDR - Daily AI Papers
SCOPE: Synthetic Conditional Objectives for Policy Evolution in Black-Box Combinatorial Optimization
・Black-box combinatorial optimization requires systematically identifying high-quality solutions under a limited evaluation budget, yet the unknown objective function provides little guidance for deciding where the search should explore next.
・We introduce SCOPE, a general framework for Synthetic Conditional Objectives for Policy Evolution in Black-Box Combinatorial Optimization.
・Rather than directly optimizing the ina
cs.LG updates on arXiv.org
SDO: Structure-Aware Data Organization for Efficient LLM Post-Training
・arXiv:2607.27273v1 Announce Type: new Abstract: Post-training of large language models is expensive, and existing efficiency improvements mainly focus on selecting informative samples or designing training schedules.
・However, data organization itself is usually treated as a static preprocessing step: embedding-based grouping methods construct fixed partitions before training and cannot adapt to the evolving sample ex
cs.LG updates on arXiv.org
SE(3)-MeanFlow: Few-Step Protein Backbone Generation on Lie Groups
・arXiv:2607.27431v1 Announce Type: new Abstract: Generative modeling of protein backbones promises the de novo design of proteins with prescribed structural and functional properties.
・Existing diffusion and flow-matching models produce high-quality backbones on SE(3)^N, but inference requires numerically integrating an ODE over hundreds of network evaluations, each involving a Lie group exponential map - a bottleneck
cs.LG updates on arXiv.org
Search Strategies for Optimal Classification and Regression Trees
・arXiv:2607.28170v1 Announce Type: new Abstract: Optimal decision trees (ODTs) are compact, interpretable machine learning models that globally optimize a given objective, but their scalability remains challenging.
・While recent work has proposed a variety of search strategies to improve scalability, the precise contribution of each strategy remains unclear.
・To address this gap, we introduce a general algorithmic frame
cs.LG updates on arXiv.org
Secure Aggregation for Privacy-Preserving Federated Learning on Clinical EEG Data
・arXiv:2607.28191v1 Announce Type: cross Abstract: Federated learning enables multiple institutions to train shared models without exchanging raw clinical EEG data, but it does not fully prevent privacy leakage from individual model updates.
・This paper presents a privacy-preserving federated learning framework for clinical EEG data using masking-based secure aggregation as the core protection mechanism.
・The framework
cs.LG updates on arXiv.org
Selecting Open-Weight Language Models for Zero-Shot Intent Classification: A Systematic Evaluation of 41 Models
・arXiv:2607.27421v1 Announce Type: cross Abstract: Intent classification is a core component of task-oriented dialogue systems, yet practitioners have limited systematic guidance for selecting deployable open-weight language models under compute, latency, and robustness constraints.
・We present a systematic zero-shot evaluation of 41 open-weight language models spanning 15 families and the 135M--9B parameter range acro
cs.LG updates on arXiv.org
Semi-Supervised Learning for Molecular Graphs via Ensemble Consensus
・arXiv:2607.28304v1 Announce Type: new Abstract: Machine learning is transforming molecular sciences by accelerating property prediction, simulation, and the discovery of new molecules and materials.
・Acquiring labeled data in these domains is often costly and time-consuming, whereas large collections of unlabeled molecular data are readily available.
・Standard semi-supervised learning methods often rely on label-preser
cs.LG updates on arXiv.org
ShadowDancer: Teaching Video World Models Any Action by Learning Unified Dynamics Representations from a Video and Its Shadow
・arXiv:2607.28362v1 Announce Type: cross Abstract: We present ShadowDancer, a novel approach to any-action, frame-level control of interactive video world models.
・The obstacle is representational: existing interfaces either encode an action loosely, leaving how it unfolds for the model to improvise, or encode it exactly through structured signals that serve one family and are hard to acquire, so precise control across
Takara TLDR - Daily AI Papers
Share the Judge, Learn the Deferral: Where Specialization Helps LLM Evaluation
・Agentic systems have widened the gap between producing candidate outputs and reviewing them.
・This paper asks a practical architectural question: should domain specialization be built into an evaluator's weights, or into the rule that decides when its judgment can be trusted?
・We study 99,952 public, rubric-conditioned examples.
stat.ML updates on arXiv.org
SiamJEPA: On the Role of Siamese Student Encoders in JEPA
・arXiv:2607.04044v2 Announce Type: replace-cross Abstract: Recently, Joint Embedding Predictive Architectures (JEPAs) have attracted significant attention in the computer vision and machine learning communities as a promising framework for self-supervised representation learning.
・Unlike masked autoencoders that reconstruct pixels, JEPA models learn representations by predicting latent embeddings of masked regions.
cs.LG updates on arXiv.org
SNAC-Pack 2.0: Scaled-Out Surrogate Neural Architecture Codesign
・arXiv:2605.16138v3 Announce Type: replace Abstract: Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost.
・This gap is particularly large for FPGA deployment, where cost is dominated by a multi-dimensional budget of lookup tables, DSPs, f
cs.LG updates on arXiv.org
Sparsity Induced Identifiability in Matrix Tri-Factorisation
・arXiv:2607.27507v1 Announce Type: new Abstract: Matrix factorisation is a fundamental tool for exploiting low-dimensional structure in high-dimensional data, with applications such as data compression, denoising, structure discovery, interpretable representation learning, and dimensionality reduction.
・Compared to conventional two-factor models, matrix tri-factorisation provides greater modelling flexibility, while sp
cs.LG updates on arXiv.org
SpecPrefetch: Parameter-Efficient Expert Prefetching for Sparse MoE Foundation Models
・arXiv:2607.24787v2 Announce Type: replace-cross Abstract: Sparse Mixture-of-Experts (MoE) models expand foundation model capacity through conditional expert activation, but their full expert pools remain difficult to deploy under limited accelerator memory.
・Although expert offloading alleviates memory pressure by moving inactive experts to host memory or storage, it introduces a routing-dependent transfer bottleneck:
Takara TLDR - Daily AI Papers
SPFM-Net: Semantic-Prior-Guided Frequency-Constrained Mamba for Invisible Watermark Attack
・Existing watermark attacks typically rely on predefined signal-processing operations or locally constrained restoration networks, making it difficult to capture the long-range dependencies of globally distributed watermark signals and resulting in an unfavorable trade-off between removal effectiveness and visual fidelity.
・In this paper, we propose SPFM-Net, a semantic-prior-guided and frequency-constrained Mamba fram
cs.LG updates on arXiv.org
Stage-Replay Divergence Follows the KV Cache: Fixed-Prefix Precision Controls and Bidirectional Cache Transplantation
・arXiv:2607.28495v1 Announce Type: new Abstract: Stage-replay diagnostics reconstruct intermediate token prefixes and treat fresh-prefill continuation as continuation from the decoder state that originally reached the prefix.
・We audit that assumption at a whole reasoning-stage boundary in a Qwen2.5-derived system.
・A matched 200-item experiment compares retained live cache with one-shot prefill of identical integer tok
cs.LG updates on arXiv.org
STEREODISCO: Discovering Stereotypicality in LLMs
・arXiv:2607.27824v1 Announce Type: cross Abstract: LLMs encode, convey, and perpetuate stereotypes.
・Prior computational research focuses on a small set of semantic axes investigated in social psychology, and operates on word embeddings produced by language models, leaving open which other semantic axes carry stereotypical associations in LLMs and how LLMs internally represent such axes.
・We introduce STEREODISCO, a fra
cs.LG updates on arXiv.org
Strategies for Milestone-driven Start-ups in Multi-activity Settings
・arXiv:2607.27563v1 Announce Type: new Abstract: New venture start-ups need to ``survive'' through multiple stages of reaching milestone targets.
・We investigate the strategies for start-ups in a milestone-oriented setting.
・We examine a model of an entrepreneurial start-up firm, where its state is captured by a diffusion process.
cs.LG updates on arXiv.org
Strategy, Not Payoffs: A Behavioural Embedding of Normal-Form Games
・arXiv:2607.27536v1 Announce Type: cross Abstract: Learning a strategic task changes more than what is directly taught: fine-tuning on one game can either enhance or degrade an agent's ability to reason in another.
・Understanding and predicting this transfer of strategic capabilities, however, remains a key challenge for large language models (LLMs).
・Normal-form games provide an ideal testbed for analyzing this phenome
cs.LG updates on arXiv.org
Subtract or Replay? Exact Deletion from Language-Model Memory
・arXiv:2607.27539v1 Announce Type: new Abstract: Exact deletion from persistent language-model memory depends on how that memory represents a record.
・Addressable influence can be removed by algebraic decrement; influence transformed by later writes inside shared recurrent state requires rebuilding from before the write.
・We test this distinction in two pretrained models against explicit record-omitted references.
cs.LG updates on arXiv.org
Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups
・arXiv:2607.27232v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly shaping how we consume information and form our worldview.
・This raises concerns beyond bias in AI: do LLMs grasp the emotional nuances conveyed via textual framing?
・In this work, we empirically evaluate how well an array of LLMs aligns with human emotional perception.
cs.LG updates on arXiv.org
TAPO: Transition-Aware Policy Optimization for LLM Agents
・arXiv:2607.27973v1 Announce Type: new Abstract: Recently, Reinforcement Learning (RL) has emerged as a crucial paradigm for the post-training of Large Language Model (LLM) agents.
・However, existing methods predominantly rely on sparse task rewards for policy optimization, failing to fully exploit another class of inherently dense supervisory signals naturally present during online interaction: environmental feedback
cs.LG updates on arXiv.org
The Confidence Manifold: Geometric Structure of Correctness Representations in Language Models
・arXiv:2602.08159v2 Announce Type: replace Abstract: When a language model asserts that "the capital of Australia is Sydney," does it know this is wrong?
・Models assert misconceptions with the same fluency as facts, so the question cannot be answered from output uncertainty.
・Truth-related signals are known to exist in the residual stream, but not their geometry: how many dimensions carry the signal, how simple a detect
cs.LG updates on arXiv.org
The Convergence Behavior of Adam under Heavy-Tailed Noise
・arXiv:2607.27383v1 Announce Type: new Abstract: We establish the first convergence guarantees for the plain vector-form \emph{Adam} optimizer under heavy-tailed stochastic noise.
・While several Adam variants are known to achieve optimal iteration complexity in bounded-variance nonconvex optimization, little is understood about their behavior when stochastic gradients admit only a bounded $p$-th central moment for some
MachineLearningMastery.com
The End-to-End Agentic AI Pipeline
・In this article, you will learn the seven architectural components that separate a production-grade agentic AI system from a demo script, and how each one...
Takara TLDR - Daily AI Papers
The Geometric Nature and a Free Proxy for Flow-Matching Uncertainty
・Flow matching (FM) has become a popular action head paradigm for modern embodied models.
・However, as a conditional generative model, it does not explicitly expose its inherent uncertainty, producing faulty action chunks even when it misinterprets the scene or encounters out-of-distribution (OOD) inputs.
・Therefore, determining when an FM-generated action can be trusted is essential for safe deployment, yet existing un
cs.LG updates on arXiv.org
The Kinetics of Training: A Driven-Nucleation Rate Law for Emergence, Plasticity Loss, and Circuit Control in Language Models
・arXiv:2607.27281v1 Announce Type: new Abstract: A capability appears in a language model when the last parts of its circuit align in one stochastic attempt, and getting all but one right is worth nothing.
・We show this no-partial-credit joint alignment is the rate-limiting step of capability formation.
・Two fingerprints: in a shortcut-free apparatus a five-part circuit missing three waits as long as a three-part circui
cs.LG updates on arXiv.org
The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence
・arXiv:2605.26494v2 Announce Type: replace-cross Abstract: We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence.
・The flagship M2 contains 229.9B total parameters with only 9.8B activated per token.
・Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-drive
stat.ML updates on arXiv.org
The Phase Transition in Online PCA Depends on $n/d\log(d)$, not $n/d$
・arXiv:2607.23914v2 Announce Type: replace-cross Abstract: High dimensional statistical theory has established the importance of constant aspect ratio, when the number of dimensions ($d$) and samples ($n$) satisfy $n,d\to\infty$ with $n/d\to \gamma\in(0,\infty)$, in understanding the limits of canonical estimation problems.
・In particular, for estimating the top eigenvector of a $d\times d$ population covariance matrix
cs.LG updates on arXiv.org
The Role of Causality in Algorithmic Recourse
・arXiv:2607.28497v1 Announce Type: new Abstract: Algorithmic recourse aims to provide individuals with actionable changes to improve their predicted outcomes in high-stakes classification settings, such as loan and mortgage applications.
・However, most existing approaches focus only on flipping a model's prediction, without accounting for whether the recommended changes lead to genuine improvement in an individual's tr
cs.LG updates on arXiv.org
The Topological Trouble With Transformers
・arXiv:2604.17121v4 Announce Type: replace Abstract: Transformers encode structure in sequences via an expanding contextual history.
・However, their purely feedforward architecture fundamentally limits dynamic state tracking.
・State tracking -- the iterative updating of latent variables reflecting an evolving environment -- involves inherently sequential dependencies that feedforward networks struggle to maintain.
cs.LG updates on arXiv.org
Theatre Chapbooks At Scale: A Statistical Comparative Analysis of Typography
・arXiv:2607.27266v1 Announce Type: cross Abstract: We propose a statistical methodology that quantifies the similarity of typefaces between printed historical books.
・This provides a tool that accelerates philological analysis.
・Using character prototypes derived from clustering and aligning automatically extracted character images, the method defines a typeface distance between any two books.
cs.LG updates on arXiv.org
THGFM: Dual-Branch Temporal Heterogeneous Graph Fusion Model
・arXiv:2607.27303v1 Announce Type: new Abstract: Temporal heterogeneous graphs offer a natural abstraction for dynamic relational systems in which diverse node and relation types co-exist and evolve over time.
・Learning on such graphs requires jointly modeling cross-type structural heterogeneity and the temporal dynamics of interactions, yet existing methods still struggle to reconcile parameter-efficient cross-type tr
cs.LG updates on arXiv.org
TIER-MoE: Trust-Informed Expert Routing via Conditional Modality Risk for Multimodal Fusion in Biomedical Classification
・arXiv:2607.27289v1 Announce Type: new Abstract: The promise of multimodal fusion lies in combining complementary sources of evidence, yet more evidence does not always yield a better prediction.
・Recent multimodal models have advanced fusion through richer cross-modal interaction and sample-adaptive fusion.
・However, the influence assigned to a modality during fusion does not reveal whether that source is unreliable, r
cs.LG updates on arXiv.org
Tight Bounds for Learning Polyhedra with a Margin
・arXiv:2604.14614v2 Announce Type: replace-cross Abstract: We give an algorithm for PAC learning intersections of $k$ halfspaces with a $\rho$ margin to within error $\varepsilon$ that runs in time $\textsf{poly}(k, \varepsilon^{-1}, \rho^{-1}) \cdot \exp \left(O(\sqrt{n \log(1/\rho) \log k})\right)$.
・Notably, this improves on prior work which had an exponential dependence on either $k$ or $\rho^{-1}$ and matches know
cs.LG updates on arXiv.org
Tight Sample Complexity for Low-Rank Adaptation: Matching Bounds and Rank Selection
・arXiv:2607.27680v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has become the standard mechanism for fine-tuning large pretrained models, yet its statistical properties remain only partially understood.
・Existing generalization results provide upper bounds of the form O~(sqrt(rd/n)) or O~(rd/n), but a matching lower bound is missing, and the question of how to choose the LoRA rank r has no formal answer.
cs.LG updates on arXiv.org
TopoFormer: Topology Meets Attention for Graph Learning
・arXiv:2607.28259v1 Announce Type: new Abstract: We introduce Topoformer, a lightweight and scalable framework for graph representation learning that encodes topological structure into attention-friendly sequences.
・At the core of our method is Topo-Scan, a novel module that decomposes a graph into a short, ordered sequence of topological tokens by slicing over node or edge filtrations.
・These sequences capture multi-sc
cs.LG updates on arXiv.org
Topological Data Analysis combined with Machine Learning for Predicting Permeability of Porous Media
・arXiv:2605.17581v2 Announce Type: replace-cross Abstract: Flow in porous media is difficult to address using standard analytical or numerical methods due to its complexity.
・However, since synthetic representations of porous media are easy to produce and data from physical experiments are becoming more widely available, the problem is well-suited to studies that include machine learning (ML) techniques.
・We discuss a n
cs.LG updates on arXiv.org
Towards joint scaling laws with optimal batch size schedules
・arXiv:2607.27731v1 Announce Type: new Abstract: Modern deep learning typically keeps the batch size static throughout training, thus overlooking the joint effect of learning rate and batch size on the training dynamics.
・In this paper, we study the deep learning dynamics through the lens of convex optimization and derive a joint characterization of loss in terms of both schedules, applicable to general optimizers and
Takara TLDR - Daily AI Papers
Towards Practical Algorithm Selection for Unsupervised Domain Adaptation in Medical Imaging
・Numerous unsupervised domain adaptation (UDA) algori-thms exist, but for clinical practice, selecting the best-suited one along with proper hyperparameters often remains unclear, as the unlabeled deployment (target) domain prevents direct evaluation.
・We propose a label-free criterion that jointly selects the algorithm and hyperparameters for UDA.
・Given a pool of candidate models from multiple algorithms trained with
cs.LG updates on arXiv.org
Towards Stability of Parameter-Free Optimization
・arXiv:2405.04376v4 Announce Type: replace Abstract: Hyperparameter tuning, particularly the selection of an appropriate learning rate in adaptive gradient training methods, remains a challenge.
・To address this challenge, we propose a novel parameter-free optimizer, \textsc{AdamG} (Adam with the Golden step size), designed to automatically adapt to diverse optimization problems without task-specific learning-rate tuni
cs.LG updates on arXiv.org
Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization
・arXiv:2607.27767v1 Announce Type: new Abstract: Graph neural networks (GNNs) can operate on large graphs but become infrastructure-sensitive at the scale of millions of nodes and typically require scalable training techniques for even larger graphs.
・This raises a central question: when can a model trained on a smaller, scaled-down replica of a graph be deployed on the full-resolution graph without retraining?
cs.LG updates on arXiv.org
Transporting Task Vectors across Different Architectures without Training
・arXiv:2602.12952v3 Announce Type: replace Abstract: Adapting large pre-trained models to downstream tasks often produces task-specific parameter updates that are expensive to relearn for every model variant.
・While recent work has shown that such updates can be transferred between models with identical architectures, transferring them across models of different widths remains unexplored.
・In this work, we introduce The
cs.LG updates on arXiv.org
TriShield: Zero-Utility-Loss Defense Against Privacy Backdoors in Federated Language Model Fine-Tuning via Orthogonal Gradient Projection and Optimizer State Entanglement
・arXiv:2607.27940v1 Announce Type: new Abstract: Federated fine-tuning of large language models (LLMs) enables collaborative training without exposing raw data.
・However, a recent attack, NeuroImprint [1] (arXiv:2606.20553), demonstrates that a malicious parameter server can corrupt a PEFT adapter into a privacy backdoor: by assigning a dedicated memorization neuron to each training sample and ensuring each neuron upda
Takara TLDR - Daily AI Papers
TriShield: Zero-Utility-Loss Defense Against Privacy Backdoors in Federated Language Model Fine-Tuning via Orthogonal Gradient Projection and Optimizer State Entanglement
・Federated fine-tuning of large language models (LLMs) enables collaborative training without exposing raw data.
・However, a recent attack, NeuroImprint [1] (arXiv:2606.20553), demonstrates that a malicious parameter server can corrupt a PEFT adapter into a privacy backdoor: by assigning a dedicated memorization neuron to each training sample and ensuring each neuron updates at most once, the server can analytically re
stat.ML updates on arXiv.org
Uncertainty in Physics and AI: Taxonomy, Quantification, and Validation
・arXiv:2605.10378v2 Announce Type: replace Abstract: Reliable uncertainty quantification is essential for the use of machine learning in physics, where scientific discoveries depend on validated probabilistic statements.
・We provide a structured overview of uncertainty quantification in ML for physics, introducing a unified taxonomy of uncertainty and clarifying the interpretation of predictive and inference uncertaint
cs.LG updates on arXiv.org
Uncertainty quantification for trustworthy deep learning: Methods and measures
・arXiv:2607.28248v1 Announce Type: cross Abstract: The deployment of deep neural networks in safety-critical domains demands reliable estimates of predictive confidence, yet conventional architectures lack principled uncertainty quantification.
・This survey provides a structured, critical review of methods for Uncertainty Quantification (UQ) in deep learning, scoped to ensemble-based and approximate Bayesian approaches
cs.LG updates on arXiv.org
Understanding Submodular Information Measure Based Objectives for Representation Learning: A Variance and Separation Perspective
・arXiv:2607.27660v1 Announce Type: new Abstract: Submodular Information Measures (SIMs) have recently emerged as a powerful framework for representation learning and multimodal learning.
・In particular, the SCORE framework~\cite{majee2024score} demonstrated that SIMs can serve as effective objectives for supervised contrastive learning.
・Despite their empirical success, however, the geometric and statistical properties
Takara TLDR - Daily AI Papers
Understanding Submodular Information Measure Based Objectives for Representation Learning: A Variance and Separation Perspective
・Submodular Information Measures (SIMs) have recently emerged as a powerful framework for representation learning and multimodal learning.
・In particular, the SCORE framework~\cite{majee2024score} demonstrated that SIMs can serve as effective objectives for supervised contrastive learning.
・Despite their empirical success, however, the geometric and statistical properties induced by different submodular information meas
OpenAI News
Univé builds an AI-ready workforce
・See how Univé built an AI-ready workforce with ChatGPT Enterprise by combining leadership, responsible governance, and employee-led innovation to transform work at scale.
cs.LG updates on arXiv.org
Variance-Aware Baselines and Adaptive Learning Rates for Reinforcement Learning with Verifiable Rewards
・arXiv:2511.23310v3 Announce Type: replace-cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as an effective paradigm for post-training large language models, yet the design of its baselines and learning-rate schedules remains largely heuristic.
・This limits our understanding of the statistical properties of policy-gradient estimators and their interaction with optimization dynamics.
cs.LG updates on arXiv.org
VESTIGE: A Knowledge-Guided Masking Strategy for Corruption-Aware Fine-Tuning of Genomic Transformers, Validated on Ancient DNA Reconstruction
・arXiv:2607.27712v1 Announce Type: new Abstract: Standard masked-language-model fine-tuning applies a uniform masking probability across every token position, assuming reconstruction difficulty is position-agnostic.
・When the degradation process is characterised and concentrated at predictable positions, this assumption fails: at peak damage sites the model can underperform a frequency-matched random predictor.
Takara TLDR - Daily AI Papers
ViP-Rig: Visual-Prompted Controllable Rigging
・Rigging is inherently task-dependent because the same mesh may require different skeletons and deformation behaviors across animation tasks.
・In practice, artists often inspect an initial rig and repeatedly edit its skeletal structure and deformation behavior to meet specific animation requirements.
・Existing automatic methods primarily generate a plausible rig from geometry, offering limited explicit control over the
cs.LG updates on arXiv.org
Weather Emulators at the Frontier of Heat Extremes Predictability
・arXiv:2607.28220v1 Announce Type: cross Abstract: Atmospheric predictability declines rapidly beyond the next ten days, such that forecasts at longer lead times primarily convey large-scale trends rather than specific states.
・Yet in a warming world, improving early warnings of extreme heat is an increasingly critical challenge.
・Here we evaluate six state-of-the-art deep learning weather emulators - Pangu-Weather, FuX
Takara TLDR - Daily AI Papers
Weather Emulators at the Frontier of Heat Extremes Predictability
・Atmospheric predictability declines rapidly beyond the next ten days, such that forecasts at longer lead times primarily convey large-scale trends rather than specific states.
・Yet in a warming world, improving early warnings of extreme heat is an increasingly critical challenge.
・Here we evaluate six state-of-the-art deep learning weather emulators - Pangu-Weather, FuXi, ArchesWeather, AIFS, GraphCast and Aurora - alo
cs.LG updates on arXiv.org
What Is The Performance Ceiling of My Classifier? Utilizing Category-Wise Influence Functions for Pareto Frontier Analysis
・arXiv:2510.03950v2 Announce Type: replace Abstract: Data-centric learning seeks to improve model performance from the perspective of data quality, and has been drawing increasing attention in the machine learning community.
・Among its key tools, influence functions provide a powerful framework to quantify the impact of individual training samples on model predictions, enabling practitioners to identify detrimental sam
cs.LG updates on arXiv.org
What Makes Deep Learning Work for Traditional Chinese Medicine Tongue Diagnosis? A Comprehensive Ablation Study
・arXiv:2607.28148v1 Announce Type: cross Abstract: Deep learning has shown promise for automated tongue diagnosis in traditional Chinese medicine (TCM), yet the design space remains underexplored.
・We conducted a systematic ablation study spanning 20+ model versions under rigorous 5-fold cross-validation on TongueDx2 (5,109 images, 976 expert-annotated) and a merged dataset of 11,101 samples.
・We compared six backbone a
cs.LG updates on arXiv.org
What Makes Graph Unified? Principles and Generative Sliding-Window Transformer for Graph Foundation Models
・arXiv:2607.27966v1 Announce Type: new Abstract: Graph Foundation Models (GFMs) have recently emerged as a promising paradigm for general-purpose graph learning, aiming to learn reusable knowledge that generalizes across diverse graph domains and downstream tasks, reducing the need for specific model development.
・Achieving this goal requires reconciling the substantial heterogeneity in node features, graph structures,
Takara TLDR - Daily AI Papers
What Makes Graph Unified? Principles and Generative Sliding-Window Transformer for Graph Foundation Models
・Graph Foundation Models (GFMs) have recently emerged as a promising paradigm for general-purpose graph learning, aiming to learn reusable knowledge that generalizes across diverse graph domains and downstream tasks, reducing the need for specific model development.
・Achieving this goal requires reconciling the substantial heterogeneity in node features, graph structures, and semantic information across domains.
cs.LG updates on arXiv.org
What Must a Fairness Audit Report When Demographic Data Is Incomplete?
・arXiv:2506.23033v5 Announce Type: replace Abstract: Fairness audits are a key component of responsible machine-learning deployment.
・Yet what such an audit must disclose, when the protected labels it depends on are incomplete, remains unsettled.
・In this work, we focused on the rates a fairness audit publishes and on what an oversight reader needs beside them.
cs.LG updates on arXiv.org
When Does Explicit View Routing Work? A Controlled Study of Multi-View Graph-Text Alignment
・arXiv:2607.27530v1 Announce Type: new Abstract: Graph-text retrieval typically maps a graph and its description to a single embedding, even when a query concerns only one semantic aspect, such as a class label or molecular property.
・Multiple heads can separate these aspects, but a change in the query head may alter retrieval even when the wrong text is sent to that head.
・Such behavior demonstrates architectural chann
cs.LG updates on arXiv.org
When unlearning is free: leveraging low influence points to reduce computational costs
・arXiv:2512.05254v2 Announce Type: replace Abstract: As concerns around data privacy in machine learning grow, the ability to unlearn, or remove, specific data points from trained models becomes increasingly important.
・While state of the art unlearning methods have emerged in response, they typically treat all points in the forget set equally.
・In this work, we challenge this approach by asking whether points that have
cs.LG updates on arXiv.org
Why Are GUI Agents Correct but Late? Decode on the Decision-Time Critical Path, Tested with Pre-Compiled Policy Trees
・arXiv:2607.28399v1 Announce Type: new Abstract: Computer-use agents often fail on transient GUI events because they produce the correct action only after the relevant window has already closed.
・We identify the main cause as expensive autoregressive decoding on the decision-time critical path.
・We propose Adaptive Anticipatory Policy Trees (AAPT), which eliminates this delay without modifying the underlying model.
cs.LG updates on arXiv.org
WIDE: Boosting Adaptive LLM Inference via Token-level Dynamic Width Pruning
・arXiv:2607.28418v1 Announce Type: cross Abstract: Pruning is a promising approach for improving the efficiency of LLMs.
・Existing static structured pruning methods are hardware-friendly and can deliver practical throughput gains, but their input-agnostic computation allocation often causes substantial accuracy degradation under aggressive sparsity.
・Recent dynamic sparsity methods improve quality retention by adapting
cs.LG updates on arXiv.org
Windowed thinning and query complexity for the bouncy particle and Zigzag samplers
・arXiv:2607.28413v1 Announce Type: cross Abstract: Let $\mu(d x)\propto e^{-U(x)} d x$ on $\R^d$, where $U$ is $m$-strongly convex and $L$-smooth, and denote by $\kappa=L/m$ the condition number.
・We consider windowed thinning, an exact simulation method for the bouncy particle sampler and the coordinate Zigzag process.
・The method divides a trajectory into deterministic windows and uses a gradient evaluation at the beg
cs.LG updates on arXiv.org
Wiring diagram extraction and gluing: a case study in classifying figure skating jumps using 3D dataset
・arXiv:2607.27598v1 Announce Type: cross Abstract: Hasse clustering is an algorithm that extracts common patterns in sequential data and represents them in graphical forms.
・As the number of expected clusters grows, however, the algorithm can become infeasible to run due to combinatorial complexity.
・In this article, we describe a theory of gluing wiring diagrams, allowing iterative applications of Hasse clustering to a
cs.LG updates on arXiv.org
ZAPs: A Reward Attribution Framework for DeFi Ecosystems with Adversarial-Robust Scoring via Parallel Anomaly Ensemble Detection
・arXiv:2607.27859v1 Announce Type: cross Abstract: Incentive programs are central to user acquisition in decentralized finance, but many reward systems rely on raw volume, transaction count, and wallet count, making them vulnerable to bots and sybil operations.
・We present ZAPs, a reward attribution framework that combines economic contribution scoring with adversarial robustness.
・A composite activity score uses protoc
Takara TLDR - Daily AI Papers
ZAPs: A Reward Attribution Framework for DeFi Ecosystems with Adversarial-Robust Scoring via Parallel Anomaly Ensemble Detection
・Incentive programs are central to user acquisition in decentralized finance, but many reward systems rely on raw volume, transaction count, and wallet count, making them vulnerable to bots and sybil operations.
・We present ZAPs, a reward attribution framework that combines economic contribution scoring with adversarial robustness.
・A composite activity score uses protocol-specific percentile normalization to limit whal
cs.LG updates on arXiv.org
ZUNA1.1: A more flexible EEG foundation model for Denoising and Super-resolution
・arXiv:2607.27308v1 Announce Type: new Abstract: We introduce ZUNA1.1, a 380M-parameter diffusion autoencoder for flexible EEG signal reconstruction.
・ZUNA1.1 is capable of reconstructing variable length sequences of up to 30s, with an arbitrary number of EEG channels at arbitrary scalp locations, and can reconstruct arbitrary temporal intervals within channels in addition to reconstructing entire channels.
＠IT 全フォーラム 最新記事一覧
アイリスオーヤマも悩んだ「無線LANがつながりにくい」問題 何を変えて解決した？
・Web会議の利用拡大で、無線LANによる安定した通信の確保が課題となっていたアイリスオーヤマ。同社は全国24拠点の無線LANをどのように見直し、通信品質を改善したのか。
ITmedia NEWS 最新記事一覧
アスクル、個人情報の漏えい懸念が約134万件に 新たに約60万件特定
・アスクルは7月30日、2025年10月に受けたランサムウェア攻撃に伴う障害について、外部への漏えいのおそれがある個人情報約60万件を追加で特定したと発表した。氏名、住所、電話番号、メールアドレスが含まれる。現時点で、外部への漏えいの事実や不正利用は確認していないという。
ITmedia NEWS 最新記事一覧
セブン＆アイ、共通会員IDのPayPay統合を正式発表 ソフトバンクや三井住友カードなどが計3000億円出資
・セブン＆アイ・ホールディングスとセブン-イレブン・ジャパンは7月31日、ソフトバンク、PayPay、LINEヤフー、三井住友カードとの戦略的パートナーシップを発表した。セブン＆アイの共通会員ID「7iD」を「PayPay ID」に統合する。また、ソフトバンクや三井住友カードなどと資本業務提携も結び、各社から1000億円ずつ、総額3000億円の出資を受ける。
ITmedia NEWS 最新記事一覧
ダイナミックマッププラットフォーム、「3Dmapspocket」の一般道点群データの提供エリアを拡大
・ダイナミックマッププラットフォームは30日、3D空間プラットフォーム「3Dmapspocket」において、一般道点群データの提供エリア拡大を発表した。関東・中部・東北地方の11都県のデータ利用可能範囲を拡充した。
ITmedia NEWS 最新記事一覧
タカラトミー、デュエマアプリで個人情報漏えいか 最大15万5000人分 氏名や住所など閲覧の恐れ
・タカラトミーは7月28日、スマートフォンアプリ「デュエル・マスターズ サポートアプリ」で最大約15万5000人分の個人情報を第三者が閲覧できる状態だったと発表した。原因はユーザー認証機能の脆弱性で、アプリの公開から11カ月余り続いていた。
＠IT 全フォーラム 最新記事一覧
ランサム対応が機能しない企業の共通点 危険なボトルネックを解消するには？
・「身代金は絶対に支払うべきではない」という原則だけではランサムウェア対応で答えが出ないこともあります。では、有事の混乱の中で難しい判断に迫られないために企業は平時に何を決めておくべきでしょうか。ボトルネックとそれを解消するポイントを解説します。
ITmedia NEWS 最新記事一覧
ロシアの「.ru」ドメインが実質使えなくなりそう？ 本人確認義務化でユーザー悲鳴、関連事業者は対応急ぐ
・ロシア政府が「.ru」ドメインなどの管理者に本人確認を義務付ける方針を示し、同ドメインに関連する事業者が対応を迫られている。SNSでは、対象のドメインを使うユーザーから不安の声も出ている。
Zennの「大規模言語モデル」のフィード
開発業務におけるAI活用ベストプラクティス（2026年版）
・この記事の対象読者 開発業務で日常的にAIを活用している人 AIに作業を委譲したいが、どのようなやり方が良いのかわからない人 1.
・なぜ「プロンプトだけ」では足りないのか 2022〜2024年は Prompt Engineering（指示の書き方）が中心でした。しかし本番運用では、次のような限界が露呈しています。
・1回の指示では複数ステップの作業を完遂できない モデルが「完了しました」と言っても、実際には未検証 コンテキストが増えると精度が落ちる ツール呼び出しの失敗・ループ・権限逸脱が起きる その結果、業界の焦点は Prompt → Context → Harness → ...
ITmedia NEWS 最新記事一覧
熊本地震の被害推定、衛星データ基に地図化 現地の航空写真400枚超も公開 国際航業
熊本地震の被害推定、衛星データ基に地図化 現地の航空写真400枚超も公開 国際航業
Zennの「大規模言語モデル」のフィード
生成は雑でいい、検証器が全部拾う：採択率0.26%で成立する『生成→検証』ループ
・🎯 はじめに LLMを実務に組み込むほど、「たぶん合ってる」出力をどう扱うかに悩まされる。
・コード生成にせよ、SQL生成にせよ、構造化データにせよ、 「モデルが確信を持って返してくる誤り」を人間側で拾い切るのは現実的ではない。
・このとき効くパターンが、生成と検証を分離するという考え方だ。
ITmedia NEWS 最新記事一覧
全国の「歴史の古い草原」分布を初めて地図化──シンク・ネイチャー
・シンク・ネイチャーのサイエンスチームは、100年以上にわたって人為的な管理が継続されてきた「歴史の古い半自然草原（古草原）」の分布を地図化し、閲覧用のWebサイトを公開した。
＠IT 全フォーラム 最新記事一覧
日立はMythosをどこまで使いこなしたのか？ 実証で見えたリアルな性能
・日立製作所はClaude Mythos Previewを使った脆弱性の特定と修正に関する実証成果を発表した。日立グループ内の100件を超えるユースケースに適用した結果、同社はMythosをどう評価したのか。リアルな性能を紹介する。
ITmedia NEWS 最新記事一覧
入門機だけど「これがいい」 XREALの「xbx a01+」は62gしかない大画面だった
・XREALから登場した「xbx a01+」は、日常のさまざまなシーンで自分だけの大画面を得られるARグラス。実際に使い倒してみると、そこにはスペックからは分かりにくい使い勝手の良さがありました。