ai Trend Report
Dashboard へ戻る
3
High impact
5
Mid impact
366
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。
Takara TLDR - Daily AI Papers
frb100-40 After Two Decades: An Optimality Certificate and a Preregistered Search Study
・For more than 20 years, the Model-RB benchmark frb100-40 remained an open challenge; since 2014, its public record had stood at 99 of 100 variables.
・We give a directly checkable 100-vertex independent set for its 4,000-vertex graph.
・Together with a verified partition into 100 cliques of size 40, the witness proves that the maximum independent-set size is 100 and the minimum vertex-cover size is 3,900.
＠IT 全フォーラム 最新記事一覧
HDDは「古いほど壊れる」とも限らない？ モデル別の監視で見えた“故障要因”
・HDDは長く使うほど故障しやすくなるのか。Backblazeが34万台超のHDDを対象にした2026年第1四半期のドライブ統計を公開した。長期間使われているHDDから新たに導入された大容量HDDまで、その故障率を分析すると、HDDの故障要因を巡る興味深い実態が見えてきた。
cs.LG updates on arXiv.org
Refining Heuristic-Based Bitcoin Address Clustering with Graph Neural Networks
・arXiv:2609.01942v1 Announce Type: new Abstract: Bitcoin's pseudonymous nature makes it challenging to analyze user-level activity, since a single user may control multiple identifiers (addresses).
・Existing heuristic-based methods attempt to identify addresses belonging to the same user, but they often produce flat cluster assignments with limited modularity and are prone to errors such as merging different users toge
cs.LG updates on arXiv.org
Adaptive Graph-of-Islands Evolution for Automatic Feature Engineering with LLMs
・arXiv:2607.23286v2 Announce Type: replace-cross Abstract: Automatic feature engineering (AutoFE) for tabular data requires discovering informative transformations from a large program space.
・Existing approaches suffer from three limitations: classical methods rely on fixed operator libraries with limited expressivity, LLM-based methods generate proposals from static prompts without retaining search experience, and ev
cs.LG updates on arXiv.org
Enhancing brain age estimation with structural MRI and synthesized cerebral blood volume maps
・arXiv:2412.01865v5 Announce Type: replace-cross Abstract: BrainAGE is a promising imaging-derived biomarker of neurobiological ageing and disease risk, yet current approaches rely predominantly on T1-weighted structural MRI, overlooking functional vascular changes that may precede tissue damage and cognitive decline.
・DeepCBV maps, synthesized from non-contrast MRI, offer a scalable alternative to contrast-enhanced pe
cs.LG updates on arXiv.org
Gauge dependence and structured-output corruption in sign-branched repetition penalties: measurements across models, inference stacks, and alternative repetition controls
・arXiv:2607.09791v2 Announce Type: replace Abstract: The multiplicative repetition penalty shipped across the LLM inference ecosystem (HuggingFace, vLLM, llama$.$cpp, and a dozen further engines) branches on the sign of each raw logit (divide positives by theta, multiply negatives).
・But the softmax is unchanged by adding a constant to every logit, so a model's logit zero-point is arbitrary (a gauge choice), and the si
cs.LG updates on arXiv.org
Hybrid Retrieval-Augmented Generation with Knowledge Graph Expansion, RRF Fusion, and Per-Chunk Grounded Evaluation for Enterprise Document Search
・arXiv:2609.01617v1 Announce Type: cross Abstract: Getting accurate, grounded answers out of large enterprise document repositories is a difficult problem.
・Dense vector retrieval alone frequently performs poorly on queries that mix technical terminology, vendor-specific acronyms, or require reasoning across several non-adjacent sections.
・DocuSearch was built to address exactly this gap - an offline, multi-agent docume
cs.LG updates on arXiv.org
Medical Heuristic Learning: An LLM-Driven Framework for Interpretable and Auditable Clinical Decision Rules
・arXiv:2606.16337v4 Announce Type: replace-cross Abstract: Predictive modeling for clinical decision support requires both strong predictive performance and transparent, auditable, and human-reviewable decision logic.
・Although deep learning and tree-based ensemble methods can achieve high accuracy, their black-box nature remains a major obstacle to trustworthy clinical deployment.
・Moreover, clinical prediction often o
NVIDIA Blog
‘NBA 2K27’ With NVIDIA DLSS 5 Leads 28 New Games Coming to GeForce NOW
・September is here with 28 more games streaming on GeForce NOW this month, led by a slam dunk: NBA 2K27 with the NVIDIA DLSS 5 3D-Guided Neural Rendering feature.
・Through NVIDIA’s close collaboration with Visual Concepts and 2K, DLSS 5 brings a new level of lifelike lighting and material detail to the court — tuned […]
＠IT 全フォーラム 最新記事一覧
“取りあえず再起動”実は危険？ 「PCが動かない」と言われたときの起動トラブル対処法
・従業員から「PCが起動しない」との報告を受けたとき、情報システム担当者はどう対処すればよいのか。原因によって確認の仕方も、対処方法も異なる。いざというときに迷わないためのポイントを見ていこう。
ITmedia NEWS 最新記事一覧
“純金金貨”購入→中身は銅と鉄 田中貴金属やミキモトかたる偽サイトに注意 消費者庁
・消費者庁は9月3日、「田中貴金属工業」や「ミキモト」をかたって貴金属や宝飾品の偽物を販売する偽サイトについて注意喚起した。SNS広告から誘導し、代金引換で偽物を送り付ける手口だという。
ITmedia NEWS 最新記事一覧
「AI導入したのに、仕事が楽にならないんだが？」の正体 虚空に消えるあなたの“削減時間”、その行き先は……
・AI活用に伴う「仕事が楽にならない」の声。得られたはずの削減効果はどこに消えているのか……。
＠IT 全フォーラム 最新記事一覧
「DRAMもSSDも高い」いつまで続く？ メモリ争奪戦で見えてきた“出口の違い”
・PCやサーバのメインメモリとして使われる「DRAM」や、ストレージに使われる「NAND型フラッシュメモリ」の需給が逼迫（ひっぱく）する状況が続いている。今後も価格上昇は続くのか。メモリ争奪戦はいつまで続く見通しなのか。
ITmedia NEWS 最新記事一覧
「Yahoo!スマホ買取」開始 オンライン査定＋配送で完結 ヤフオクで販売検討
・買い取ったスマホは、「Yahoo!オークション」「Yahoo!ショッピング」を通じた販売を検討する。
ITmedia NEWS 最新記事一覧
「アメリカ湖」を拒否したらApp Storeでランキング1位に “忘れられていた”老舗地図アプリとは？
・米国のドナルド・トランプ大統領が8月27日に署名した「オンタリオ湖」を「アメリカ湖」へ名称を変更する大統領令が、地図サービスを提供するIT企業にも影響を与えている。中でも注目を集めたのが、表記変更の拒否を表明した地図アプリ「MapQuest」だ。
＠IT 全フォーラム 最新記事一覧
「ついにペーパー廃止」 情報処理技術者試験が令和8年度からCBT方式に全面移行、何が変わる？
・IPAが令和8年度における応用情報技術者試験、高度試験および情報処理安全確保支援士試験の申込受付期間や試験実施期間、試験科目の名称変更、成績発表の仕組み、受験申し込み手順、試験会場での注意事項などを発表した。令和8年度前期試験の申し込みページも公開した。
＠IT 全フォーラム 最新記事一覧
「ネットワークがもはや時代遅れ」 IT部門が疲弊するクラウド接続、どう解消？
・企業はクラウドやAI利用を支えるためのネットワークの準備ができていると考える傾向にある。だがIXを運営するDE-CIXの調査では、それとは異なる実態が浮かび上がった。企業のIT部門が疲弊しているクラウド接続問題と、それを解消する方法とは。
＠IT 全フォーラム 最新記事一覧
「パスワードは頭で記憶」が4割超、パスワードの“使いまわし”が横行する理由
・利用するWebサービスが増えるほど、パスワードの管理は難しくなる。安全性を考えればサービスごとに異なるパスワードを設定したいところだが、現実にはそう簡単ではないようだ。300人を対象にした調査から、パスワード管理の実態が明らかになった。
ITmedia NEWS 最新記事一覧
「違反見つけてもSNSで晒さないで」 ホロライブ、二次創作ガイドライン改訂
・「本ガイドライン違反と考えられる行為を見つけた場合であっても、これを理由として、他のファンや参加者を攻撃・非難したり、SNS等で晒すなどの行為はお控えください」
＠IT 全フォーラム 最新記事一覧
「強制アップデートは必須」 モバイルアプリ開発で事前に知っておきたいポイント10選
・モバイルアプリ開発で事前に確認すべきことを整理できていますか？数十万ユーザー規模のアプリをReact Nativeで開発・運用する過程で分かった「気づき・教訓」をもとに、開発時に知っておきたいスマートフォン固有機能の仕様確認から、運用時に知っておきたい証明書更新まで幅広く取り扱います。
＠IT 全フォーラム 最新記事一覧
「君のCLAUDE.mdはもう盛り過ぎ」 Claude 5世代で“通用しなくなった”コンテキストの6つの常識
・Anthropicは公式ブログで、「Claude 5」世代モデルにおけるコンテキストエンジニアリングの新しいベストプラクティスを解説した。
ITmedia NEWS 最新記事一覧
「好き嫌い.com」ってどんなサイト？ 運営は個人、国が情プラ法で指定した“好感度掲示板”を開いてみた
・総務省は8月31日、有名人への「好き」「嫌い」を匿名で投票できる「好き嫌い.com」を運営する個人を、情報流通プラットフォーム対処法に基づく大規模事業者に指定した。個人運営のサイトが対象に入るのは初めて。そもそもどんなサイトなのか、実際に開いて確かめた。
ITmedia NEWS 最新記事一覧
「仏にドラゴンボールのテーマパーク」報道に東映アニメ「許諾していない」……どこで話がねじくれた？ 話題が錯綜した経緯
・フランスに「ドラゴンボール」のテーマパークができる――国内外で報じられた計画を巡り、東映アニメーションは8月31日、「一切のライセンス許諾を行っていない」との声明を発表した。日本では、当初の報道が「なぜ日本で立ち上げられなかった」といった議論の種になっていただけに、同社の声明は「許可なしで話が進んでいたのか」と驚きと共に受け止められている。
＠IT 全フォーラム 最新記事一覧
【Pythonで学ぶデータ分析】相関があるかどうかをベイズ統計で調べる ～ 年齢と原付事故死傷者数に関係はあるのか？
・バイクに乗る若い人の運転はどうも危なっかしく感じられることがあります。そこで、手軽に乗れる原付一種（かつての排気量50ccまでのバイク）を対象に、年齢と交通事故死傷者数との間に関係があるかどうかを調べてみたいと思います。つまり、相関のベイズ検定に取り組もうというわけです。『社会人1年生から学ぶやさしいデータ分析』ベイズ統計編の第8回です。
＠IT 全フォーラム 最新記事一覧
1社100万円で非エンジニアをAI人材に ホリエモンAI学校が法人貸切プラン
・ホリエモンAI学校は生成AIを利用した業務改善やシステム開発を学ぶ研修プログラムにおいて、新たに企業単位での貸切プランの提供を開始した。条件を満たす中小企業の場合、助成金を活用できる可能性があるという。
cs.LG updates on arXiv.org
A Common Measure of Communication for Speech Brain-Computer Interfaces
・arXiv:2609.02887v1 Announce Type: new Abstract: Speech brain-computer interfaces (speech BCIs) translate neural activity into language, offering a path towards restoring speech for people with paralysis and, more broadly, enabling new forms of natural human-computer interaction.
・Despite this promise, the field lacks a common measure of progress because systems use different datasets, recording methods, types of speec
cs.LG updates on arXiv.org
A Comparative Study of Graph Representations for GNN-Based Power Grid Control in L2RPN
・arXiv:2609.02538v1 Announce Type: new Abstract: Graph construction is a critical but underexamined design choice in deep reinforcement learning for power grid control.
・We present a controlled experimental comparison of different graph representations, including physical topology, electrical-sensitivity, and hybrid variants for topology control in the Learning to Run a Power Network (L2RPN) environment.
・Our findings i
cs.LG updates on arXiv.org
A computational approach to maximum likelihood thresholds for colored Gaussian graphical models
・arXiv:2609.02382v1 Announce Type: cross Abstract: Gaussian graphical models (GGMs) are essential tools for interpretable structure learning.
・However, in high-dimensional, small-sample regimes, the available data is often insufficient for the maximum likelihood estimator to exist.
・Colored Gaussian graphical models (CGGMs) mitigate this limitation by imposing symmetry constraints through graph coloring, which reduces t
cs.LG updates on arXiv.org
A Computational Comparison of Fourier Spectral Differentiation and Spatial Automatic Differentiation in Periodic Physics-Informed Neural Networks
・arXiv:2609.02110v1 Announce Type: new Abstract: Physics-informed neural networks (PINNs) commonly evaluate the spatial derivatives appearing in partial differential equation residuals using automatic differentiation (AD), whose computational and memory costs can become substantial when multiple or high-order derivatives are required.
・We perform a controlled comparison of spatial AD and Fourier spectral differentiatio
The latest research from Google
A connectomics milestone: Mapping the complete male fruit fly brain
A connectomics milestone: Mapping the complete male fruit fly brain
cs.LG updates on arXiv.org
A Geometry-Aware Triplane Field Network for Vehicle Aerodynamic Prediction
・arXiv:2606.07724v2 Announce Type: replace Abstract: High-fidelity computational fluid dynamics (CFD) is crucial to vehicle aerodynamic analysis, but its cost still constrains early-stage design exploration.
・Machine-learning-based surface-field prediction offers a faster alternative if the model can efficiently capture both global flow context and local geometric detail.
・This work proposes a machine-learning-based met
cs.LG updates on arXiv.org
A Multivariate Bernoulli-Based Sampling Method for Multi-Label Data with Application to Meta-Research
・arXiv:2512.08371v5 Announce Type: replace Abstract: Datasets may contain observations with multiple labels.
・If the labels are not mutually exclusive, and if the labels vary greatly in frequency, obtaining a sample that includes sufficient observations with scarcer labels to make inferences about those labels, and which deviates from the population frequencies in a known manner, creates challenges.
・In this paper, we c
cs.LG updates on arXiv.org
A Power Law in Logarithm's Clothing: On the Scalability of Graph-Based Vector Search
・arXiv:2609.02143v1 Announce Type: cross Abstract: Most vector databases rely on graph-based indexes, notably HNSW and Vamana, for approximate nearest neighbor search.
・With embedding models widely adopted, the datasets these databases store grow rapidly.
・At a fixed accuracy, how does search cost scale with dataset size?
cs.LG updates on arXiv.org
A Study of Conditional Diffusion Models for Open-Loop Control under Dry Friction and Stiction
・arXiv:2609.01756v1 Announce Type: new Abstract: Diffusion models have recently emerged as expressive generative priors for planning and control.
・This paper studies Action Diffusion, an action-sequence diffusion formulation used as an open-loop proposal distribution for a point-mass system with dry friction and stiction.
・In this benchmark, motion starts only when the applied input exceeds a static-friction threshold,
cs.LG updates on arXiv.org
A Survey on Self-Improving Test-Time Intelligence: Feedback-Driven Adapting, Learning, and Scaling at Inference
・arXiv:2609.01679v1 Announce Type: new Abstract: The ability of AI systems to improve their behavior during deployment is becoming increasingly important.
・As inference moves beyond the static execution of a fixed trained model, a growing body of work studies how models can refine their behavior on the fly by exploiting test-time information and additional computation.
・These developments have largely evolved along two
cs.LG updates on arXiv.org
A Unified Particle Filter LSTM for Data-Driven Process Simulation
・arXiv:2609.01967v1 Announce Type: new Abstract: Data-driven process simulation aims to generate realistic case trajectories from historical event logs without requiring an explicitly specified model of the underlying dynamics.
・Deep sequence models can capture complex temporal dependencies through next-activity probabilities and conditional time distributions.
・However, event logs provide only a partial view of the und
cs.LG updates on arXiv.org
A Unified Rate-Distortion Perspective on Vector, Product, and Scalar Quantization
・arXiv:2609.02107v1 Announce Type: new Abstract: Discrete visual tokenization, predominantly driven by vector, scalar, and product quantization, lacks a unified conceptual framework for understanding quantization tradeoffs.
・In this paper, we propose a unified rate--distortion perspective on modern discrete visual tokenization.
・By viewing quantization as lossy compression, we characterize the nominal fixed-length codin
cs.LG updates on arXiv.org
Achieving More with Less: A Tensor-Optimization-Powered Ensemble Method
・arXiv:2408.02936v3 Announce Type: replace Abstract: Ensemble learning is a method that leverages weak learners to produce a strong learner.
・However, obtaining a large number of base learners requires substantial time and computational resources.
・Therefore, it is meaningful to study how to achieve the performance typically obtained with many base learners using only a few.
cs.LG updates on arXiv.org
Act More, Decide Less: Skill-Guided Adaptive Action Chunking for Long-Horizon LLM Agents
・arXiv:2609.02042v1 Announce Type: new Abstract: Large language model (LLM) agents for long-horizon interactive tasks typically follow a ReAct-style protocol, issuing one primitive action per LLM round.
・While this enables frequent replanning, it is inefficient for long-horizon tasks where many rounds are spent on routine action sequences.
・A natural alternative is to let the agent emit variable-length action chunks.
cs.LG updates on arXiv.org
Action abstractions for amortized sampling
・arXiv:2410.15184v2 Announce Type: replace Abstract: As trajectories sampled by policies used by reinforcement learning (RL) and generative flow networks (GFlowNets) grow longer, credit assignment and exploration become more challenging, and the long planning horizon hinders mode discovery and generalization.
・The challenge is particularly pronounced in entropy-seeking RL methods, such as generative flow networks, wher
stat.ML updates on arXiv.org
Adaptive Replication Strategies in Trust-Region-Based Bayesian Optimization of Stochastic Functions
・arXiv:2504.20527v3 Announce Type: replace-cross Abstract: We develop and analyze a method for stochastic simulation optimization based on Gaussian process models within a trust-region framework.
・We focus on settings where the variance of the objective function is large, making accurate estimation challenging and often requiring many evaluations.
・To address this regime, we combine local modeling with adaptive replicat
ITmedia NEWS 最新記事一覧
Adobe、Slackから「Photoshop」などのツールを使える「Adobe for Slack」提供開始 会話からPDFや動画を生成
・Adobeは、Slack上で「Photoshop」や「Firefly」など70種以上のツールを使える「Adobe for Slack」の提供を開始した。Slackbot経由で指示を出し、チャットやcanvasの内容を文脈として反映した画像や資料を制作できる。業務チャットをAIの実行基盤とする動きが加速している。
cs.LG updates on arXiv.org
Adversarial Stress Testing of Outlier Detection in Subjective Image Quality Assessment
・arXiv:2509.06554v2 Announce Type: replace-cross Abstract: In subjective image and video quality assessment, observers rate or compare selected stimuli.
・Before calculating mean opinion scores (MOSs), unreliable ratings should be identified and handled as outliers.
・Several outlier-detection methods are available, including standardized procedures, but their comparative performance is often evaluated using only specific
Takara TLDR - Daily AI Papers
AffectDelta: Beyond Emotion Labels for Image Editing
・Emotion-driven image editing aims to evoke a specified target emotion by modifying emotion-relevant visual cues in a source image, while preserving the overall composition and semantic-structural coherence of the original scene.
・Existing scene-level editors typically specify the target with a single emotion category and often learn visual transformations from operation-level text instructions.
・A category collapses a
cs.LG updates on arXiv.org
Agentic Scaffolding Amplifies Sycophantic Behavior in Large Language Models
・arXiv:2608.21377v2 Announce Type: replace-cross Abstract: Sycophancy in large language models, the tendency to prioritize user agreement over truthful responses, has been documented extensively but studied primarily in single-turn settings.
・This paper investigates a critical question: does subjecting LLMs to greater interaction scaffolding make sycophancy better or worse?
・Across 4,800 veracity judgments (200 statemen
cs.LG updates on arXiv.org
AGI Maze Prediction Datasets: A Compact Benchmark for Learning World Dynamics with Transformers
・arXiv:2609.02339v1 Announce Type: new Abstract: World modeling requires a predictive model to maintain and update an internal state adequate for reasoning about the consequences of actions.
・We introduce the AGI Maze Prediction Datasets and Benchmark, a lightweight controlled testbed for studying this capability in Transformers and other predictive models.
・Derived from procedurally generated, stateful grid worlds, the
Takara TLDR - Daily AI Papers
AGI Maze Prediction Datasets: A Compact Benchmark for Learning World Dynamics with Transformers
・World modeling requires a predictive model to maintain and update an internal state adequate for reasoning about the consequences of actions.
・We introduce the AGI Maze Prediction Datasets and Benchmark, a lightweight controlled testbed for studying this capability in Transformers and other predictive models.
・Derived from procedurally generated, stateful grid worlds, the benchmark comprises per-step transition predict
cs.LG updates on arXiv.org
AI Contextual Measurement for Recovering Individual and Group-Level Effects: Validation Against Survey Measures and an Occupational Application
・arXiv:2609.02821v1 Announce Type: cross Abstract: Researchers increasingly use artificial intelligence to construct measures of social, organizational, and occupational characteristics that are absent from conventional surveys.
・We propose AICOME, AI COntextual MEasurement, a framework for evaluating whether AI-derived respondent-level measures can recover individual and group-level effects in contextual models.
AI Weekly — AI News & Updates
AI Weekly Issue #528: What are companies building with AI? An Applied AI Deep Dive
・We went looking for what companies are actually building with AI.
・The answer was not more chatbots.
・It was drones carrying diagnostic samples, driverless Frito-Lay trucks, AI-guided flight paths, repair copilots, and rugged GPU laptops in Ukraine.
＠IT 全フォーラム 最新記事一覧
AIインフラ支出、ついに「推論」が「学習」を上回る――背景についてGartnerが解説
・GartnerがAIインフラにおける予測を公開。AIに最適化されたIaaSの世界的支出が2027年には56.5％増加し、660億ドルに達するという。
＠IT 全フォーラム 最新記事一覧
AIを理由に「新卒採用を停止」した企業が22％ Gartnerが警鐘、若手を切る企業の“将来のツケ”
・企業の最高人事責任者110人に対するGartnerの調査によると、回答者の所属企業の22％では、ビジネス部門リーダーの少なくとも1人がAIによる自動化を理由に、新規学卒者レベルの人材採用を停止している。
cs.LG updates on arXiv.org
Aletheia: An Offline-First Clinical Decision Support System for Differential Diagnosis in Low-Resource Healthcare Settings
・arXiv:2607.24814v2 Announce Type: replace-cross Abstract: Access to specialist clinical expertise remains severely limited across sub-Saharan Africa, where physician-to-patient ratios can fall below 1:25,000 in rural settings.
・Existing AI-assisted diagnostic tools predominantly require reliable internet connectivity and high-specification hardware, rendering them impractical for frontline healthcare workers in distri
Engineering at Meta
An Organizational Second Brain: Building an AI That Learns From Experts
・We’ve built an AI agent that acts as a secondary expert for a given domain, making deep specialist knowledge readily available and preserved for anyone in an organization to access, share, and build upon.
・This is not a typical domain-specific agent.
・Its novelty comes from integrating two layers: A structured, auditable knowledge architecture separates what [...] Read More...
＠IT 全フォーラム 最新記事一覧
Anthropic、Claudeの「逸脱」事例を受け安全対策を強化 報酬ハッキングの問題も指摘
・Anthropicは7月末と8月末に発生したAIモデル評価中の重大インシデントについて、初期分析と安全対策を発表した。その中で、AIモデルが評価や報酬を得るための抜け道を見つける行為を確認し、対策を取ってきたことを明らかにしている。
＠IT 全フォーラム 最新記事一覧
AnthropicがフロンティアAIセキュリティサービス「EFS」を発表 履歴データを顧客環境に保持しながら自動監視で誤用を検知
・Anthropicが、厳格なデータ保護を必要とする企業向けに、セキュリティとプライバシーを両立するサービスを発表した。入力データや利用履歴などのデータが顧客環境から出ることはない。一方で、AIの悪用を防ぐための複数セッションにまたがる監視を自動で行う。
OpenAI News
ATV Big Air Tour turned 3 days of work into 3 hours with ChatGPT
・ATV Big Air Tour uses ChatGPT Work to speed up marketing, merchandising, and more.
・It even turned merchandise photos into an inventory website in 15 minutes.
ITmedia NEWS 最新記事一覧
AWS、日米結ぶ新たな太平洋海底ケーブル「Sta'O'Nuk」発表 420Tbpsで2029年に稼働へ
AWS、日米結ぶ新たな太平洋海底ケーブル「Sta'O'Nuk」発表 420Tbpsで2029年に稼働へ
ITmedia NEWS 最新記事一覧
AWSをゲームで学べる“AWSクエスト”にまさかの新バージョン AIバーチャル顧客と対話可能に
・米Amazon Web Services（AWS）は、3DオープンワールドでAWSを学べるオンラインゲーム「AWS Cloud Quest」の新バージョン「AWS Cloud Quest 2.0」の提供を開始しました。
＠IT 全フォーラム 最新記事一覧
BacklogをCLIから操作 ヌーラボ開発者が「bee」をOSSで公開
・ヌーラボの開発者が、Backlogをターミナルから操作できるCLIツール「bee」を独自に開発し、オープンソースとして公開した。課題管理やWikiなど90超のサブコマンドを備え、AIエージェントからの操作にも対応するという。
cs.LG updates on arXiv.org
Basin Geometry and Reliable Recall of Dynamical Memories in Reservoir Computing
・arXiv:2609.01914v1 Announce Type: cross Abstract: Reliable attractor recall conventionally requires broad basins of attraction.
・However, in reservoir-computing based associative memory, temporal cues reliably recover dynamical memories despite basins dominated by unpredictable, riddled-like regions.
・We reveal that memory basins exhibit an ``octopus-like'' structure: a robust ``head'' near the attractor and thin, inte
cs.LG updates on arXiv.org
Bayes-Optimal BER and AUC: Estimation and Evaluation of Estimators
・arXiv:2609.02304v1 Announce Type: new Abstract: A fundamental quantity in machine learning is the optimal performance achievable by any model on a given task.
・Estimating this quantity allows us to distinguish the irreducible part of the error from a deficiency of the model, telling us how much room for improvement remains.
・Recent work has shown that the Bayes error, or equivalently the optimal accuracy, can be estima
cs.LG updates on arXiv.org
Beyond State Consistency: Behavior Consistency in Text-Based World Models
・arXiv:2604.13824v2 Announce Type: replace Abstract: World models have been emerging as critical components for assessing the consequences of actions generated by interactive agents in online planning and offline evaluation.
・In text-based environments, world models are typically evaluated and trained with single-step metrics such as Exact Match, aiming to improve the similarity between predicted and real-world states,
cs.LG updates on arXiv.org
Beyond Transfer Accuracy: Mechanism-Guided Controlled Adaptation for Low-Resource Languages
・arXiv:2601.08146v4 Announce Type: replace-cross Abstract: Existing circuit discovery methods rely on templated tasks with clean counterfactuals, limiting their use on diverse natural text.
・We adapt Contextual Decomposition for Transformers (CD-T) for unstructured settings via label-balanced activation means and task-directional relevance scoring, enabling counterfactual-free circuit discovery.
・We leverage the discove
Takara TLDR - Daily AI Papers
Blending Concepts: Benchmarking Visual Metaphor Generation in Text-to-Image Models
・Text-to-image (T2I) models have achieved remarkable success at faithfully rendering specified objects and attributes, yet their ability to produce visual metaphors, images that convey abstract ideas by combining elements from two distinct domains, remains largely unexamined.
・To bridge this gap, we introduce VMetaphor-Bench, the first benchmark for evaluating visual metaphor generation in T2I models.
・It comprises 1,50
cs.LG updates on arXiv.org
Breadth Beats Depth: Improving GCG-Based Jailbreak Optimization with Breadth-Oriented Suffix Search
・arXiv:2609.02172v1 Announce Type: cross Abstract: Optimization-based jailbreak attacks such as Greedy Coordinate Gradient (GCG) achieve strong effectiveness and transferability by optimizing adversarial suffixes on white-box source models.
・However, existing GCG-based methods rely on averaged adversarial loss and deep greedy search, which can over-emphasize easy-to-jailbreak behaviors and overlook promising regions of
Takara TLDR - Daily AI Papers
C$^{3}$T: Counterfactual Causal Reasoning for Sentiment Shifts in Social-Media Conversation Trees
・Sentiment in social-media threads does not only vary across posts; it shifts as users react to claims, corrections, evidence, and hostility within a branching reply tree.
・We study why sentiment changes in rumor-centric conversation trees by treating discourse moves (e.g., denial/correction, evidence/link, toxicity/attack) as candidate interventions and asking (i) what sentiment a reply expresses, (ii) whether the sen
cs.LG updates on arXiv.org
CACTUS: Mask-Guided Semantic Clean-Label Backdoors in Decentralized Federated Learning
・arXiv:2609.02450v1 Announce Type: new Abstract: Semantic triggers in federated learning (FL) can be less conspicuous than synthetic patches, but sample-dependent placement may weaken backdoor implantation across aggregation rounds.
・This challenge is compounded in decentralized FL (DFL), where topology-dependent peer aggregation repeatedly mixes local models.
・CACTUS converts label-consistent semantic pairs into target
cs.LG updates on arXiv.org
CAHR-Net: Condition-Adaptive Hysteresis Reconstruction for Compact and Interpretable Magnetic Core Loss Modeling
・arXiv:2609.01991v1 Announce Type: new Abstract: Magnetic core loss originates in the hysteresis loop: the energy dissipated per excitation cycle equals the loop area, and frequency, temperature, and waveform shape set the loss by reshaping the loop geometry.
・Most existing models let these conditions act only on a terminal scalar - empirical equations fold them into fitted exponents, and data-driven predictors append
cs.LG updates on arXiv.org
Cantelli Constrained Policy Optimization
・arXiv:2601.22993v5 Announce Type: replace Abstract: We introduce Canary, a risk-averse method designed to optimize Value-at-Risk (VaR) constrained reinforcement learning (RL) problems.
・We employ Cantelli's inequality to obtain a tractable, conservative and smooth bound on the VaR constraint based on the first two moments of the cost return.
・This yields a constraint estimator that remains stable with tight violation t
cs.LG updates on arXiv.org
CAPTURE: Disentangling Preference Drift from Memory Poisoning in Personalized LLM Agents
・arXiv:2609.02265v1 Announce Type: new Abstract: Personalized language agents use persistent memory to adapt to users over time, but the same mechanism creates an attack surface.
・When new information conflicts with stored preferences, an agent must distinguish genuine preference drift from temporary context shifts, ambiguity, or adversarial memory poisoning.
・We formulate this problem as a continuous-time partially obs
cs.LG updates on arXiv.org
CAT-Flow: Curvature-Adaptive sTeps for Flow Matching
・arXiv:2609.01746v1 Announce Type: new Abstract: Flow Matching has emerged as a leading framework for generative modeling, powering state-of-the-art systems such as FLUX and Stable Diffusion 3.5.
・However, the iterative nature of its ODE-based sampling process creates a fundamental efficiency bottleneck: the quality of generated samples is highly sensitive to the choice of step-sizes, and current models typically requi
ITmedia NEWS 最新記事一覧
Claudeの“見えない透かし”適用開始 Fable 5.1とMythos 5.1が出力する文やコード、画像に付与 日本も対象
・米Anthropicが「Claude」の出力に埋め込むと予告していた機械可読な透かしが、9月1日（現地時間）提供開始の最新モデル「Claude Fable 5.1」と「Claude Mythos 5.1」で有効になった。ファイルの来歴を確かめる無料ツールの公開も始まっている。
cs.LG updates on arXiv.org
Cliff: Learning Process Rewards from the First Mistake
・arXiv:2609.02817v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for large language model (LLM) post-training, but its reliance on coarse outcome rewards leads to limited guidance on intermediate reasoning processes.
・Existing approaches such as process reward modeling and on-policy distillation introduce additional constraints, such as reliance o
cs.LG updates on arXiv.org
CliffRank: A Dual-Branch Framework for Activity-Cliff Ranking Prediction
・arXiv:2609.01673v1 Announce Type: new Abstract: Activity-cliff ranking remains difficult because local structural changes can cause large activity differences, while high-quality data that resolve the underlying mechanisms remain limited.
・To use available activity labels more effectively, we combine absolute-activity regression with ranking-consistency learning.
・CliffRank trains two parallel predictors with mean squa
cs.LG updates on arXiv.org
Clustering Three-Way Data with Outliers
・arXiv:2310.05288v4 Announce Type: replace-cross Abstract: Matrix-variate distributions are a relatively recent addition to the model-based clustering literature, thereby making it possible to analyze data in matrix form with complex structure such as images and time series.
・Due to its recent appearance, there is limited literature on matrix-variate data, with even less on dealing with outliers in these models.
cs.LG updates on arXiv.org
Codebook Agent: Amortized Topology Design for LLM Multi-Agent Systems
・arXiv:2609.02264v1 Announce Type: cross Abstract: Adapting the communication topology of an LLM multi-agent system to each query improves both accuracy and efficiency, yet current designers treat this as conditional graph generation: a variational, autoregressive, or diffusion decoder searches the $N \times N$ adjacency space, and a graph-network proxy trained on utility and a structural cost such as edge count ranks
cs.LG updates on arXiv.org
CodePoisonRAG: Knowledge Poisoning Attacks on Retrieval-Augmented Code Generation
・arXiv:2609.02774v1 Announce Type: cross Abstract: Retrieval-Augmented Code Generation (RACG) improves LLM-based software development by retrieving external code artifacts, documentation, and patches, and incorporating them into the generation context.
・This reliance on external knowledge introduces a critical trust boundary: poisoned artifacts can influence generated code without modifying the underlying LLM.
MachineLearningMastery.com
Combining LLM Embeddings with Tabular Features in a Unified Scikit-learn Pipeline
・In this article, you will learn how to build a unified scikit-learn pipeline that combines text embeddings generated by a lightweight open-source language model with...
Takara TLDR - Daily AI Papers
CoMerge: Conflict-Driven Preference Optimization for Multi-Task Model Merging
・Model merging provides an efficient paradigm for constructing multi-task large language models (LLMs) without full model retraining, yet it remains challenged by parameter interference.
・While existing methods aim to preserve the capabilities of individual expert models and mitigate interference, they generally do not directly learn from the potentially degraded behaviors exposed by naive merging.
・In this paper, we pr
cs.LG updates on arXiv.org
Compositional Spectral Prompts for LLM-based Online Time Series Forecasting
・arXiv:2609.02093v1 Announce Type: new Abstract: To address the sequential and evolving nature of time series, the Online Time Series Forecasting (OTSF) task has been extensively studied in multiple domains.
・Existing research focuses on adapting to non-stationary environments by employing memory buffer-based retrieval strategies.
・However, we observe that such frameworks struggle with long-term adaptation and fail to g
Takara TLDR - Daily AI Papers
Compositional Spectral Prompts for LLM-based Online Time Series Forecasting
・To address the sequential and evolving nature of time series, the Online Time Series Forecasting (OTSF) task has been extensively studied in multiple domains.
・Existing research focuses on adapting to non-stationary environments by employing memory buffer-based retrieval strategies.
・However, we observe that such frameworks struggle with long-term adaptation and fail to generalize to unseen patterns.
cs.LG updates on arXiv.org
Conditional Diffusion Posterior Alignment for Sparse-View CT Reconstruction
・arXiv:2604.21960v3 Announce Type: replace-cross Abstract: Computed Tomography (CT) is a widely used imaging modality in medical and industrial applications.
・To limit radiation exposure and measurement time, there is a growing interest in sparse-view CT, where the number of projection views is significantly reduced.
・Deep neural networks have shown great promise in improving reconstruction quality in sparse-view CT, es
cs.LG updates on arXiv.org
Connections between the F\"ollmer process and the denoising diffusion probabilistic model
・arXiv:2605.18040v2 Announce Type: replace-cross Abstract: The F\"ollmer process is a Brownian motion conditioned to have a pre-specified distribution at time 1.
・This process can be interpreted as an ``augmented'' time-compressed version of the reverse stochastic differential equation (SDE) corresponding to the denoising diffusion probabilistic model (DDPM).
・While this fact has been indirectly used to analyze DDPM sam
cs.LG updates on arXiv.org
Constrained Group Relative Policy Optimization
・arXiv:2602.05863v4 Announce Type: replace Abstract: Group Relative Policy Optimization (GRPO) remains the dominant critic-free approach for fine-tuning LLMs and VLMs, but its compatibility with constrained policy optimization (e.g.
・for safety-critical domains) has not been carefully examined.
・In this work, we introduce Constrained GRPO, a Lagrangian-based extension of GRPO for constrained policy optimization.
cs.LG updates on arXiv.org
Context Inference Attacks Without Jailbreaks
・arXiv:2609.01663v1 Announce Type: cross Abstract: Agentic AI systems are increasingly deployed to process sensitive data at inference time, such as healthcare records or financial documents assembled into a hidden \emph{context} before the system answers.
・Prior work has studied privacy risks primarily through \emph{jailbreaking} attacks that induce models to directly disclose sensitive content, but has largely overlo
Takara TLDR - Daily AI Papers
Contrastive Explanations in Quantitative Bipolar Argumentation Frameworks
・Argumentation frameworks are useful tools for representing and reasoning with information in a variety of settings, e.g.
・in supplementing AI models as they perform classification tasks, with a notable benefit of providing additional explainability.
・In this paper, we introduce contrastive explanations for Quantitative Bipolar Argumentation Frameworks (QBAFs), one such formalism.
cs.LG updates on arXiv.org
Convergence Theory of Knowledge Distillation in Asynchronous P2P Gossip Learning Network
・arXiv:2609.01952v1 Announce Type: new Abstract: Decentralized, serverless learning increasingly connects devices running different architectures, where the standard tool, decentralized SGD, is undefined as models with different parameter counts cannot be averaged.
・Knowledge distillation (KD) exchanges soft predictions rather than weights and sidesteps this obstacle, yet convergence theory for fully decentralized, asy
stat.ML updates on arXiv.org
Copula Transformations for Data-Consistent Inversion
・arXiv:2609.02832v1 Announce Type: new Abstract: Data-consistent inversion (DCI) constructs probability measures whose push-forward distributions agree with observed data, while iterative data-consistent inversion (iDCI) extends this framework to generalized stochastic inverse problems by enforcing multiple push-forward constraints sequentially.
・Although iDCI avoids the direct approximation of high-dimensional joint d
Takara TLDR - Daily AI Papers
Counter-GEO-Bench: Evaluating Defenses Against Information-Distorting Generative Engine Optimization
・Generative engine optimization (GEO) enables content producers to increase the visibility of their web pages in generative search engines, but the same techniques can deliver targeted misinformation when adversaries publish ordinary-looking GEO-optimized documents that victim large language models (LLMs) retrieve and synthesize into distorted answers.
・No existing benchmark evaluates defenses against this threat under
cs.LG updates on arXiv.org
Coverage, Not Targeting: A Structural Regime in Multi-Turn Agent Credit Assignment
・arXiv:2609.02417v1 Announce Type: new Abstract: Multi-turn agentic RL increasingly treats credit assignment as a targeting problem: given a terminal verifiable reward, per-turn methods localize credit onto the turns that mattered.
・We identify the structural quantity that predicts when this is the right move, the verifier information density V_d = k/C (the fraction of an agent's C-step causal chain whose per-turn corr
cs.LG updates on arXiv.org
CRISP: Cliff-awaRe Input-adaptive Sparse Prefilling with Structural-Mass-Motivated Routing
・arXiv:2609.01925v1 Announce Type: new Abstract: The attention prefilling phase of long-context LLM inference scales quadratically, making self-attention a severe computational bottleneck.
・Traditional sparse attention methods mitigate this through fixed patterns or offline profiling, but lack the flexibility to adapt to input-dependent attention structure.
・Recent dynamic methods address this by routing heads to sparse
cs.LG updates on arXiv.org
D-FROST: Decentralized Federated pRompt-tuning via Optimal tranSporT for Non-IID and Imbalanced Data
・arXiv:2609.01802v1 Announce Type: new Abstract: Prompt tuning provides a parameter-efficient way to adapt foundation models (FMs) by freezing the pretrained backbone and updating only a small set of learnable prompts.
・This property makes prompt tuning especially suitable for decentralized federated learning (DFL), where exchanging full-model updates can be prohibitively expensive.
・However, prompt tuning in DFL introd
cs.LG updates on arXiv.org
Deep denoising autoencoder-based non-invasive blood flow detection for arteriovenous fistula
・arXiv:2306.06865v2 Announce Type: replace Abstract: Clinical guidelines underscore the importance of regularly monitoring and surveilling arteriovenous fistula (AVF) access in hemodialysis patients to promptly detect any dysfunction.
・Although phono-angiography/sound analysis overcomes the limitations of standardized AVF stenosis diagnosis tool, prior studies have depended on conventional feature extraction methods, r
cs.LG updates on arXiv.org
Deep Reinforcement Learning for Reach-Avoid-Stay Problems
・arXiv:2410.02898v3 Announce Type: replace-cross Abstract: Reach-Avoid-Stay (RAS) tasks are essential in applications where systems must safely reach a target set and remain within it under all bounded disturbances.
・Existing approaches either struggle to compute the maximal robust RAS set, the set of all states from which the RAS task is achievable, or are limited in handling general dynamic systems.
・To address these
cs.LG updates on arXiv.org
DeepAffinity: Long-Term Aspect Preference Prediction in eCommerce using Small Language Models
・arXiv:2609.02468v1 Announce Type: new Abstract: We explore predicting eCommerce user preferences for product aspects such as brand, size, and color - a task we define as Aspect Affinity.
・Solving this task improves customer understanding and enables fine-grained personalization in recommendation, search, and marketing.
・We frame Aspect Affinity as a temporal prediction task: forecasting a users future aspect choices fr
ITmedia NEWS 最新記事一覧
DeNA、新規事業子会社「BAKERU」設立 「売上100億円規模の事業をゼロから創出」へ
・売上100億円・営業利益10億円以上へと成長する事業をゼロから生み出すことを目指す。
cs.LG updates on arXiv.org
DiDrive: A Risk-Aware Hierarchical Diffusion Framework for Safe Offline Reinforcement Learning in Autonomous Driving
・arXiv:2609.01609v1 Announce Type: new Abstract: While diffusion models effectively capture multimodal behavioral priors for autonomous driving, offline reinforcement learning (RL) policies remain susceptible to distribution shift, heavy-tailed risk signals, out-of-distribution (OOD) action generation, and high-dimensional state redundancy.
・To address these challenges, we propose DiDrive, a distribution-guided offline
cs.LG updates on arXiv.org
Differentiable Electricity-Market Clearing for Gradient-Based Planning
・arXiv:2609.02646v1 Announce Type: new Abstract: Planning a large data center is difficult because a facility big enough to matter changes the electricity prices it will pay.
・Those prices are set by market clearing, a constrained optimization problem solved anew in every operating condition.
・However, simulating the market tells a planner how a candidate plan performs but not how to improve it.
cs.LG updates on arXiv.org
Dimension Dependent Correlation Gap Bounds under Restricted Independence
・arXiv:2609.02659v1 Announce Type: cross Abstract: The pairwise independent correlation gap is the ratio of the maximum expected value of a set function under arbitrary dependence to that under pairwise independence, measuring the loss from this independence restriction.
・Under mutual independence, this gap is universally bounded by $e/(e-1)$ for monotone submodular functions.
・With pairwise independence, a tighter $4/3
cs.LG updates on arXiv.org
Discriminative World Models for Web Agents
・arXiv:2609.02885v1 Announce Type: cross Abstract: Recent web agents use world models for test-time action selection by sampling candidate actions, predicting the resulting web states, and ranking them with a ranker model or a Process Reward Model (PRM).
・These world models are typically trained via supervised next-state prediction to generate fixed representations like HTML or AXTree snapshots.
・However, this objective
Takara TLDR - Daily AI Papers
Discriminative World Models for Web Agents
・Recent web agents use world models for test-time action selection by sampling candidate actions, predicting the resulting web states, and ranking them with a ranker model or a Process Reward Model (PRM).
・These world models are typically trained via supervised next-state prediction to generate fixed representations like HTML or AXTree snapshots.
・However, this objective is misaligned with the downstream ranker, which r
cs.LG updates on arXiv.org
Disease Burden over Skin Tone: Decomposing the Dermatology-AI Generalization Gap
・arXiv:2609.02111v1 Announce Type: cross Abstract: Dermatology artificial intelligence (AI) models are predominantly trained on light-skinned, cancer-focused image collections, yet they are increasingly proposed for deployment in resource-constrained settings where patients differ from training populations along two confounded axes: skin tone and disease distribution.
・We investigate whether poor generalization is prim
Takara TLDR - Daily AI Papers
DKL: Decoupled Knowledge Learning for Instruction-Tuned Language Models
・RAG has become the de facto method for incorporating new, corpus-specific knowledge into an instruction following LLM (Instruct LLM).
・Although RAG-based prompting improves factual grounding, it fails when retrieval is incorrect or incomplete, leading to hallucinations.
・Finetuning methods such as RAFT and PA-RAG enhance RAG by injecting new knowledge into the model's parameters, but require generating a massive amount
cs.LG updates on arXiv.org
DLM-One: Diffusion Language Models for One-Step Sequence Generation
・arXiv:2506.00290v2 Announce Type: replace-cross Abstract: This paper introduces DLM-One, a score-distillation-based framework for one-step sequence generation with continuous diffusion language models (DLMs).
・DLM-One eliminates iterative refinement by aligning the scores of a student model's outputs with the score function of a pretrained teacher DLM in the forward-diffused noisy space.
・We demonstrate that our framew
cs.LG updates on arXiv.org
DMRL: Document-Mediated Reinforcement Learning for Skill Optimization in Advertising Recommendation
・arXiv:2609.02170v1 Announce Type: new Abstract: Advertising recommendation requires continuously tuning complex system parameters while balancing commercial returns and user experience.
・Recent work has introduced large language models (LLMs) with skill documents to assist this labor-intensive process, but skill optimization remains largely prompt-driven, lacking a principled mechanism to attribute rewards to specific
cs.LG updates on arXiv.org
Do Large Language Models Capture the Diversity in their Training Data?
・arXiv:2609.02275v1 Announce Type: cross Abstract: Large language models are trained to model conditional distributions over text, yet it remains inadequately understood whether they capture the full diversity of plausible outputs present in their training data.
・We study this question through an information-theoretic lens by comparing the conditional entropy of model-generated outputs with that of the corresponding tr
cs.LG updates on arXiv.org
Do Tabular Foundation Models Know Physics? Contamination, Units, and the Deterministic Limit
・arXiv:2609.02766v1 Announce Type: new Abstract: Tabular foundation models (TFMs) learn to fill in tables the way language models fill in text, and tables are arguably the format in which most physical measurement arrives.
・Did they learn any physics in the process?
・They are Bayesian by construction, so the question is what their prior contains.
Takara TLDR - Daily AI Papers
Do Tabular Foundation Models Know Physics? Contamination, Units, and the Deterministic Limit
・Tabular foundation models (TFMs) learn to fill in tables the way language models fill in text, and tables are arguably the format in which most physical measurement arrives.
・Did they learn any physics in the process?
・They are Bayesian by construction, so the question is what their prior contains.
cs.LG updates on arXiv.org
DocHop: Benchmarking Out-of-domain Multi-hop Reasoning in Information-Dense Documents
・arXiv:2609.02059v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have achieved strong performance on structured visual understanding tasks such as chart and document question answering.
・However, existing benchmarks typically evaluate these domains in isolation, leaving underexplored a key capability: whether models can use textual context to determine how chart evidence should be selected, i
Takara TLDR - Daily AI Papers
Doppio: A Dataset for Contactless Weight Estimation of Falling Particles
・Measuring the mass of powder, including falling particles, is a common task in industrial applications.
・While scales are effective for static measurements, many applications require contactless sensing, where existing solutions are often costly, application-specific, and technically complex.
・In this work, we investigate computer vision as a practical alternative for contactless mass estimation.
cs.LG updates on arXiv.org
Double-Bounded Nonlinear Optimal Transport for Size Constrained Min Cut Clusterin
・arXiv:2501.18143v2 Announce Type: replace Abstract: Min cut is an important graph partitioning method.
・However, current solutions to the min cut problem suffer from slow speeds, difficulty in solving, and often converge to simple solutions.
・To address these issues, we relax the min cut problem into a double-bounded constraint and, for the first time, treat the min cut problem as a double-bounded nonlinear optimal tra
cs.LG updates on arXiv.org
Doubly Stochastic Adaptive Neighbors Clustering via the Marcus Mapping
・arXiv:2408.02932v3 Announce Type: replace Abstract: Clustering is a fundamental task in machine learning and data science, and similarity graph-based clustering is an important approach within this domain.
・Doubly stochastic symmetric similarity graphs provide numerous benefits for clustering problems and downstream tasks, yet learning such graphs remains a significant challenge.
・Marcus theorem states that a strictly
cs.LG updates on arXiv.org
Dutch Books for Language Models
・arXiv:2609.02797v1 Announce Type: cross Abstract: People increasingly use language models to support life decisions.
・Many such decisions involve a probabilistic forecast: How likely is a major life event, a natural disaster, or an economic outcome?
・Users of language models may implicitly trust that these forecasts fall out of a coherent world model.
Takara TLDR - Daily AI Papers
Dutch Books for Language Models
・People increasingly use language models to support life decisions.
・Many such decisions involve a probabilistic forecast: How likely is a major life event, a natural disaster, or an economic outcome?
・Users of language models may implicitly trust that these forecasts fall out of a coherent world model.
cs.LG updates on arXiv.org
DynaTokens: Controlling Token Dynamics for Continual Video-Language Understanding
・arXiv:2603.06662v3 Announce Type: replace-cross Abstract: Continual VideoQA with multimodal LLMs remains challenging because sequential adaptation induces task interference, while storing task-specific prompts becomes impractical as task sequences grow.
・We introduce DynaTokens, a transformer-based token generator that dynamically produces fine-tuning tokens on demand, enabling task-adaptive prompt updates through sha
cs.LG updates on arXiv.org
DynG-Diff: A State-Aware Dynamic Guidance Diffusion Framework for Probabilistic Time Series Forecasting
・arXiv:2609.02068v1 Announce Type: new Abstract: Probabilistic multivariate time series (MTS) forecasting is crucial for modeling complex dynamical systems.
・However, existing diffusion-based methods rely on task-specific conditional paradigms that lack flexibility and struggle with inherent "information heterogeneity"--the significantly varying noise levels and evolutionary patterns across variables.
・To address this,
ITmedia NEWS 最新記事一覧
dカード、dポイント還元率を1％→0.5％に引き下げ 前年に利用なければ年会費も発生
・NTTドコモは9月1日、「dカード」の通常決済で付与するdポイントの還元率を、2027年1月利用分から1％から0.5％に引き下げると発表した。年会費も条件付きで有料化する。対象は通常のdカードのみで、上位カードである「dカード GOLD U」「dカード GOLD」「dカード PLATINUM」は含まない。
cs.LG updates on arXiv.org
Efficient Context-Limited Telescope Bibliography Classification for the WASP-2025 Shared Task Using SciBERT
・arXiv:2609.01647v1 Announce Type: new Abstract: The creation of telescope bibliographies is a crucial part of assessing the scientific impact of observatories and ensuring reproducibility in astronomy.
・This task involves identifying, categorizing, and linking scientific publications that reference or use specific telescopes.
・However, this process remains largely manual and resource intensive.
cs.LG updates on arXiv.org
Eliciting ESG Preferences for Reinforcement Learning-Based Portfolio Optimization
・arXiv:2609.02677v1 Announce Type: cross Abstract: Modern portfolio management increasingly demands a balance between traditional risk-adjusted returns and strict Environmental, Social, and Governance (ESG) mandates.
・Current Reinforcement Learning (RL) approaches typically optimize for a single ESG provider, neglecting the significant divergence in rating methodologies across the industry and the unintuitive nature of
cs.LG updates on arXiv.org
Emergence of Fibrations, Compression, and Symmetry Breaking in Artificial Neural Networks
・arXiv:2609.01768v1 Announce Type: new Abstract: Artificial neural networks are often regarded as powerful yet opaque black boxes.
・Here, we demonstrate that learning in deep neural networks generates local symmetries known in graph theory as fibrations and coverings.
・We prove that covering symmetries are stable attractors of stochastic gradient descent.
cs.LG updates on arXiv.org
Emotional regulation improves deep learning-based image classification
・arXiv:2606.13081v2 Announce Type: replace Abstract: Emotion significantly influences cognition, enhancing memory and learning under certain conditions.
・Drawing on this principle, emotion-augmented deep learning investigates how affective states can improve neural network architectures and learning paradigms, achieving better generalization than non-emotional models.
・However, existing methods often rely solely on obje
cs.LG updates on arXiv.org
Enhancing Road Safety Through Multi-Camera Image Segmentation with Post-Encroachment Time Analysis
・arXiv:2511.12018v2 Announce Type: replace-cross Abstract: Traffic safety analysis at signalized intersections is essential for reducing vehicle and pedestrian collisions, yet traditional crash-based studies are limited by data sparsity and reporting latency.
・This paper presents a multi-camera computer vision framework for real-time safety assessment through Post-Encroachment Time (PET) computation, demonstrated at th
cs.LG updates on arXiv.org
Entangled Representations Amplify Collateral Damage in Unlearning
・arXiv:2609.02285v1 Announce Type: new Abstract: A long-held intuition in interpretability research is that representational entanglement, the sharing of structure between knowledge domains in a neural network, makes unlearning harder.
・While the intuition is widespread, it has never been directly tested in a controlled experiment.
・We present a way to do so: by repurposing Selective Gradient Masking (SGTM), we train a
Takara TLDR - Daily AI Papers
Entangled Representations Amplify Collateral Damage in Unlearning
・A long-held intuition in interpretability research is that representational entanglement, the sharing of structure between knowledge domains in a neural network, makes unlearning harder.
・While the intuition is widespread, it has never been directly tested in a controlled experiment.
・We present a way to do so: by repurposing Selective Gradient Masking (SGTM), we train a suite of six 254M-parameter language models on E
cs.LG updates on arXiv.org
Evidence for Shared Routing Geometry and Dynamics in Sparse Mixture-of-Experts
・arXiv:2609.02404v1 Announce Type: new Abstract: Sparse mixture-of-experts (MoE) models use an independently parameterized router at each sparse layer to select experts for every token.
・Prior work has shown that routing decisions across depth can often be predicted from earlier routing signals, suggesting that routing is not fully independent across layers.
・However, the structure behind this predictability remains unc
cs.LG updates on arXiv.org
Exact Limits of Random Projections for Preserving Geometry: Distance Recovery, Nearest-Neighbor Rankings, and Covariance Shape in Gaussian Models
・arXiv:2609.02155v1 Announce Type: new Abstract: The Johnson-Lindenstrauss (JL) lemma guarantees that a random projection of $n$ points to $m=O(\varepsilon^{-2}\log n)$ dimensions preserves pairwise squared distances within relative error $\varepsilon$ with high probability, and this dimension order is asymptotically optimal.
・In high dimensions, however, distances concentrate around a baseline while key geometric info
cs.LG updates on arXiv.org
Exchange Policy Optimization Algorithm for Semi-Infinite Safe Reinforcement Learning
・arXiv:2511.04147v2 Announce Type: replace Abstract: Safe reinforcement learning (RL) aims to optimize long-term performance while adhering to safety requirements.
・However, many practical applications involve an infinite number of constraints, forming semi-infinite safe RL (SI-safe RL).
・Such scenarios typically appear when safety conditions must be enforced across an entire continuous parameter space, such as ensuring
cs.LG updates on arXiv.org
Explainable Information Processing in Particle Swarm Optimization through Landscape and Search Behavior Analysis
・arXiv:2509.06272v5 Announce Type: replace-cross Abstract: Swarm-based optimization algorithms have demonstrated remarkable success in solving complex problems, yet their widespread adoption remains limited due to poor transparency in how algorithmic components influence performance.
・This work presents a multi-faceted explainability framework for Particle Swarm Optimization (PSO) through two complementary perspectives
cs.LG updates on arXiv.org
FairLens: Benchmarking Fairness in Vision-Language Models for High-Stakes Decision-Making
・arXiv:2609.01691v1 Announce Type: cross Abstract: Vision-language models (VLMs) are increasingly used to make decisions from visual inputs.
・We introduce FAIRLENS, a benchmark and evaluation framework for measuring both the fairness and the validity of VLM responses in three high-stakes domains: hiring, legal, and healthcare.
・FAIRLENS pairs real face images spanning gender, race, and age groups with closed- and open-e
cs.LG updates on arXiv.org
Feature Interaction Modeling for Neural Operators
・arXiv:2607.28762v2 Announce Type: replace Abstract: Despite the many variants of DeepONet that have been proposed, query-based operator networks still struggle with shock-dominated and low-viscosity PDEs, whose sharp moving discontinuities and slowly decaying solution spectra challenge finite-dimensional separable representations.
・In this work, we propose \emph{Feature Interaction Modeling Operator} (FM-Operator), a
cs.LG updates on arXiv.org
Federated LoRA Adaptation of BiomedCLIP Across Four International Chest X-Ray Cohorts
・arXiv:2609.02101v1 Announce Type: new Abstract: Federated learning (FL) lets institutions train a shared model without exchanging data, and Low-Rank Adaptation (LoRA) makes this practical at scale by communicating only compact low-rank updates.
・Biomedical imaging is a compelling setting for this combination: patient data are archived behind privacy regulations, and institutions differ widely in scanners, protocols, a
cs.LG updates on arXiv.org
FlashKAN: B-Spline KANs via Truncated Power Form
・arXiv:2609.01956v1 Announce Type: new Abstract: Kolmogorov-Arnold Networks (KANs) place learnable B-spline activations on network edges rather than fixed activations on nodes.
・The standard Cox-de Boor recursion evaluates these activations through k sequential passes for degree-k splines, consuming over 90% of forward-pass time.
・FlashKAN replaces this recursion with the truncated power form, a classical result from ap
cs.LG updates on arXiv.org
FORGE: Forward-Only Test-Time Adaptation for Integer-Only Vision Models on Microcontrollers
・arXiv:2609.01683v1 Announce Type: cross Abstract: Vision models deployed on microcontrollers (MCUs) are quantized to integer-only arithmetic and run in inference-only runtimes that do not carry the machinery backpropagation needs: the standard tool for adapting a model to the distribution shift (sensor noise, blur, lighting) it meets in the field.
・Existing forward-only test-time adaptation (TTA) methods either run on
Takara TLDR - Daily AI Papers
From Detection to Characterization: A Large-Scale Study of Ragebait on Japanese X
・Ragebait refers to online content intentionally designed to provoke anger or outrage and thereby increase attention and engagement.
・However, reliable large-scale detection and systematic analysis of ragebait remain limited, hindering efforts to understand its prevalence, impact, and mitigation.
・This study aims to develop an effective ragebait detection framework and to clarify the characteristics of ragebait at scale
cs.LG updates on arXiv.org
From Digital to Physical Reservoir Computing: Co-Optimizing Soft Robotic Reservoirs via Dynamics Matching
・arXiv:2608.00484v2 Announce Type: replace-cross Abstract: Soft robotic substrates are promising for Physical Reservoir Computing (PRC) because their compliant nonlinear dynamics can provide temporal memory, high-dimensional state transformations, and efficient inference.
・However, physical reservoirs are often adopted as-is rather than pretrained or co-optimized, potentially limiting soft robotic PRC performance relat
cs.LG updates on arXiv.org
From High-Dimensional Spaces to Verifiable ODD Coverage for Safety-Critical AI-based Systems
・arXiv:2604.02198v2 Announce Type: replace-cross Abstract: While Artificial Intelligence (AI) offers transformative potential for operational performance, its deployment in safety-critical domains such as aviation requires strict adherence to rigorous certification standards.
・Current EASA guidelines mandate demonstrating complete coverage of the AI/ML constituent's Operational Design Domain (ODD) -- a requirement that
MIT News - Artificial intelligence
From MIT to IBM, expediting AI and quantum deployment
・MIT affiliates engage with the MIT-IBM Computing Research Lab to bring rigorous theory to production systems.
cs.LG updates on arXiv.org
From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution
・arXiv:2609.02771v1 Announce Type: cross Abstract: Training data attribution (TDA) aims to identify training examples that shape model behavior, but its intervention value depends on both which examples are selected and how they are modified.
・Influence functions (IF) estimate behavioral changes under infinitesimal reweighting, yet IF-selected examples often show limited advantages over random selection under conventio
cs.LG updates on arXiv.org
From topology learning to graph generation: A unifying perspective
・arXiv:2609.02286v1 Announce Type: cross Abstract: Learning graph structures from data is a fundamental problem that spans a wide range of signal processing and machine learning tasks.
・While significant effort has been made to tackle the problem, existing research has largely evolved along two parallel directions.
・The first seeks to infer the topology of an individual graph from observations supported on it, whereas t
cs.LG updates on arXiv.org
Full-Model Optimality for Tunable Linear Generative Priors in Compressed Sensing
・arXiv:2609.02790v1 Announce Type: cross Abstract: Generative models have been studied experimentally and theoretically as priors for inverse problems such as compressed sensing.
・Recent work by Gunn et al.
・studied the use of generative priors with tunable complexity, where a family of generative priors with varying complexity is maintained and a specific complexity can be selected at inversion time.
cs.LG updates on arXiv.org
GenCAR: Generative Counterfactual Alignment with Risk-Controlled Selection for Out-of-Distribution Recommendation
・arXiv:2609.02162v1 Announce Type: cross Abstract: Serving useful recommendations under distribution shift is crucial for balancing utility and risk in out-of-distribution (OOD) recommendation.
・However, most existing OOD methods improve ranking or construct counterfactual candidates without controlling the proxy-label false discovery rate (FDR) of the served set.
・In this work, we formulate OOD serving as the $\alpha$-
cs.LG updates on arXiv.org
General Demographic Pre-trained Models for Enhancing Predictive Performance Across Diseases and Population
・arXiv:2509.07330v3 Announce Type: replace Abstract: Foundation models for healthcare require balancing robust generalization across heterogeneous clinical populations and disease settings with the architectural simplicity needed for deployment.
・We present a pre-trained model focused on demographic attributes that enhances feature utility across medical domains in a plug-and-play fashion.
・We introduce the General Demo
cs.LG updates on arXiv.org
Generalized Regret Analysis of Thompson Sampling using Fractional Posteriors
・arXiv:2309.06349v2 Announce Type: replace-cross Abstract: Thompson sampling (TS) is one of the most popular and earliest algorithms to solve stochastic multi-armed bandit problems.
・We consider a variant of TS, named $\alpha$-TS, where we use a fractional or $\alpha$-posterior ($\alpha\in(0,1)$) instead of the standard posterior distribution.
・To compute an $\alpha$-posterior, the likelihood in the definition of the st
Takara TLDR - Daily AI Papers
Generating Medical Image Counterfactuals using Causal Explanations
・Deep learning models have achieved impressive performance in medical image diagnosis, yet their deployment in clinical settings remains constrained by limited explainability.
・Counterfactual images provide one means of auditing model behavior by showing how an image would need to change for a classifier to produce a different prediction.
・Existing approaches typically generate such explanations using auxiliary models,
cs.LG updates on arXiv.org
Generative Diffusion Surrogates with Analytical Variance Schedule
・arXiv:2609.01705v1 Announce Type: new Abstract: Stochastic transport describes physical systems in which an initially structured distribution spreads under unresolved forcing, scattering, or heterogeneous media.
・Useful surrogates for such systems should be probabilistic, time-resolved, and able to represent non-Gaussian distributional structure.
・Generative diffusion models, which corrupt data with Gaussian noise and
cs.LG updates on arXiv.org
GeoSPRINT: Geometric Redundancy-Aware Step Pruning for Inference in Diffusion Trajectories
・arXiv:2609.02160v1 Announce Type: new Abstract: Diffusion models achieve high sample quality but remain expensive at inference time because sampling requires many sequential neural function evaluations (NFEs).
・Existing acceleration methods either use fixed step-skipping schedules, adapt step sizes based on local numerical error, or require additional training.
・We introduce GeoSPRINT (Geometric Step Pruning for Infere
cs.LG updates on arXiv.org
GMTRouter: Personalized LLM Router over Multi-turn User Interactions
・arXiv:2511.08590v2 Announce Type: replace-cross Abstract: Large Language Model (LLM) routing has demonstrated strong capability in balancing response quality with computational cost.
・As users exhibit diverse preferences, personalization has attracted increasing attention in LLM routing, since even identical queries may require different models to generate responses tailored to individual needs.
・However, existing appr
cs.LG updates on arXiv.org
GONE: Structural Knowledge Unlearning via Neighborhood-Expanded Distribution Shaping
・arXiv:2603.12275v2 Announce Type: replace-cross Abstract: Unlearning knowledge is a pressing and challenging task in Large Language Models (LLMs) because of their unprecedented capability to memorize and digest training data at scale, raising more significant issues regarding safety, privacy, and intellectual property.
・However, existing works, including parameter editing, fine-tuning, and distillation-based methods,
cs.LG updates on arXiv.org
GPTBIAS: A Comprehensive Framework for Evaluating Bias in Large Language Models
・arXiv:2312.06315v2 Announce Type: replace-cross Abstract: Warning: This paper contains content that may be offensive or upsetting.
・There has been a significant increase in the usage of large language models (LLMs) in various applications, both in their original form and through fine-tuned adaptations.
・As a result, LLMs have gained popularity and are being widely adopted by a large user community.
cs.LG updates on arXiv.org
Gradient Descent on Logistic Regression with Non-Separable Data and Large Step Sizes
・arXiv:2406.05033v3 Announce Type: replace Abstract: We study gradient descent (GD) dynamics on logistic regression problems with large, constant step sizes.
・For linearly-separable data, it is known that GD converges to the minimizer with arbitrarily large step sizes, a property which no longer holds when the problem is not separable.
・In fact, the behaviour can be much more complex -- a sequence of period-doubling bif
cs.LG updates on arXiv.org
Gradient Prediction with Control Variates in the Cheap-Forward Regime
・arXiv:2511.05187v2 Announce Type: replace Abstract: We study whether otherwise-idle inference resources could reduce the scarce-GPU cost of training.
・Our analysis uses a simulated compute ledger in which fleet work is billed at a fraction of a scarce-GPU forward; all experiments run on a regular GPU.
・Our algorithm predicts gradients with a reduced-precision, inference-style reverse-mode program and combines many pred
cs.LG updates on arXiv.org
GRADSOLVE: fast exact gradients for ODE ensembles on GPUs
・arXiv:2609.02876v1 Announce Type: cross Abstract: Ordinary differential equations (ODEs) underlie models in science and engineering, and many applications need derivatives of their solutions with respect to parameters.
・Ensembles of independent trajectories suit graphics processing units (GPUs), but current GPU software forces a trade-off: the fastest ensemble solvers cannot be differentiated in reverse mode at the sp
cs.LG updates on arXiv.org
Graph Machine: Towards Better Pretraining via Edges
・arXiv:2609.02881v1 Announce Type: new Abstract: We introduce the Graph Machine (GM), an architecture that maintains an $O(n)$-sized state and accesses it through sparse, dynamic routing.
・Unlike methods with fixed-size states or sparse but static routing, GM preserves $O(n)$ complexity in its sparse layers without restricting the potentially accessible state size to $O(1)$.
・Instead, GM uses edges - pointer-like object
cs.LG updates on arXiv.org
H3DNAS: Hardware-Aware ONNX-Native 3D Point Cloud Model Compression
・arXiv:2609.02684v1 Announce Type: new Abstract: Deploying 3D point cloud models on edge hardware such as the NVIDIA Jetson Orin Nano is severely constrained by compute and memory budgets.
・Existing compression methods require access to the model's original source code, rendering them inapplicable to the Open Neural Network Exchange (ONNX) binaries commonly distributed by vendors and model repositories.
・We present \tex
cs.LG updates on arXiv.org
Half-Truth Audio Detection and Localisation: A Lightweight Cross-Attentive Architecture and a Cross-Corpus Diagnostic Study
・arXiv:2605.29531v3 Announce Type: replace-cross Abstract: Partially manipulated (half-truth) speech, where a short synthesised segment is spliced into an otherwise genuine utterance, is a harder and more realistic forensic threat than the fully synthesised deepfakes that dominate the literature.
・We present CAFNet, a lightweight (576K-parameter, 2.24 MB) cross-attentive architecture that fuses MFCC, LFCC, and Chroma-S
cs.LG updates on arXiv.org
Hardware-Accelerated Instance Segmentation for Resource-Constrained Space Robotics with Criticality Analysis
・arXiv:2609.02219v1 Announce Type: cross Abstract: Autonomous lunar missions require real-time per- ception under three coupled constraints: extreme low-light conditions, limited onboard compute, and radiation-induced hardware faults that can silently corrupt inference.
・We present a deployment-oriented instance segmentation framework for resource-constrained lunar robotics that jointly addresses quan- tization calibra
cs.LG updates on arXiv.org
Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives
・arXiv:2609.01736v1 Announce Type: cross Abstract: Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks.
・However, existing approaches suffer from two key challenges: brittle multi-step and multi-turn reasoning caused by incompatible tool output types and API schemas, and performance degradation under large tool catalogues.
・To address the
Takara TLDR - Daily AI Papers
HeadWiseKV: Budgeted Per-Head Cache Residency for Hybrid Long-Context Language Models
・Long-context inference retains a growing key--value (KV) cache during decoding, which consumes substantial GPU memory and can reduce generation throughput.
・This bottleneck remains in hybrid language models because their residual global-attention layers can dominate context-dependent cache demand.
・We study how to allocate this state under an aggregate KV-residency budget.
cs.LG updates on arXiv.org
Hearing the Whispers: Black-Box Membership Inference Attacks on Finetuned TTS Models
・arXiv:2609.01723v1 Announce Type: cross Abstract: Text-to-Speech (TTS) foundation models are increasingly fine-tuned on private datasets to synthesize highly personalized voices, introducing severe privacy risks by exposing both biometric identities and sensitive speech content.
・Existing black-box membership inference attacks (MIAs) follow a two-stage pipeline of query generation and representation engineering, both
cs.LG updates on arXiv.org
Held-out evidence resolves follow-up measurement decisions in biological screens
・arXiv:2607.27651v3 Announce Type: replace Abstract: Machine learning determines which follow-up measurements biological screens collect.
・In a six-rule Cell Painting battery, the highest-value rule would re-image 96.01% of the library and had a 97.14% false-activation upper bound, showing why predicted value alone cannot justify replacing a fixed plan.
・We developed OPAL, a held-out decision test that freezes a rule an
cs.LG updates on arXiv.org
HiPoly: a hierarchical polymer-native AI framework for property prediction and generative design
・arXiv:2609.02746v1 Announce Type: cross Abstract: Polymeric materials are central to modern technologies, with applications ranging from energy to health and transportation.
・Although AI has made significant advances in materials discovery, the hierarchical structure of polymers across multiple length scales makes them inherently difficult to represent in a unified and physically meaningful way.
・Here we introduce HiPo
cs.LG updates on arXiv.org
hLLM: Single Pass Decoding for Generative Reranking
・arXiv:2609.01807v1 Announce Type: new Abstract: Large language models (LLMs) achieve state-of-the-art generative ranking quality, but the ranking they produce must be decoded, and autoregressive decoding spends one sequential forward pass per emitted token.
・We observe that the only tokens a ranker must emit are the $N$ ordinal values naming the items in ranked order, and that this narrow, permutation-structured outpu
cs.LG updates on arXiv.org
Humanoid Safe Stop via Learned Stoppability Value
・arXiv:2609.02358v1 Announce Type: cross Abstract: Humanoid robots responding to emergency stop commands typically execute a fixed maneuver, without reasoning about whether a safe stop is actually feasible from the current state.
・We cast emergency stopping as a reach-avoid problem and propose Safe-Stop, a task-agnostic framework that pairs a learned stop policy with learned stoppability estimators.
・The estimators are
cs.LG updates on arXiv.org
HyperMC: Multi-Fidelity Hyperparameter Tuning for Stochastic Gradient MCMC
・arXiv:2609.02138v1 Announce Type: cross Abstract: Stochastic gradient Markov chain Monte Carlo (SGMCMC) methods enable scalable Bayesian inference, but their performance depends strongly on hyperparameters such as the step size, mini-batch size, and number of leapfrog steps.
・Since most SGMCMC algorithms lack a Metropolis-Hastings acceptance rate, standard acceptance-based tuning methods are not directly applicable.
cs.LG updates on arXiv.org
ICE: Intervention-Consistent Explanation Evaluation with Statistical Grounding for LLMs
・arXiv:2603.18579v2 Announce Type: replace-cross Abstract: Evaluating whether explanations faithfully reflect a model's reasoning remains an open problem.
・Existing benchmarks use single interventions without statistical testing, making it impossible to distinguish genuine faithfulness from chance-level performance.
・We show that faithfulness is not a fixed property but an operator-dependent quantity that changes with t
cs.LG updates on arXiv.org
IDEEA: training-free Input-Dependent stEEring via Activation cluster matching
・arXiv:2609.02089v1 Announce Type: cross Abstract: Steering aligns large language models (LLMs) by injecting a bias into selected activations at inference time, offering a far cheaper alternative to weight-update methods such as supervised fine-tuning or reinforcement learning.
・However, most existing training-free steering methods are input-independent: a single direction is fitted once and shared across all inputs.
Takara TLDR - Daily AI Papers
IDEEA: training-free Input-Dependent stEEring via Activation cluster matching
・Steering aligns large language models (LLMs) by injecting a bias into selected activations at inference time, offering a far cheaper alternative to weight-update methods such as supervised fine-tuning or reinforcement learning.
・However, most existing training-free steering methods are input-independent: a single direction is fitted once and shared across all inputs.
・This is fundamentally limiting as different inputs
cs.LG updates on arXiv.org
IFW-BLS: Dual-Robust Broad Learning System with Intuitionistic Fuzzy Wave Loss
・arXiv:2609.02422v1 Announce Type: new Abstract: Broad Learning System is an efficient randomized learning model that expands network width through feature and enhancement nodes and estimates the output weights without deep backpropagation.
・Its standard least-squares training, however, is vulnerable in two different ways: (i) large residuals caused by noise, outliers, or corrupted labels can dominate the objective, an
Takara TLDR - Daily AI Papers
IFW-BLS: Dual-Robust Broad Learning System with Intuitionistic Fuzzy Wave Loss
・Broad Learning System is an efficient randomized learning model that expands network width through feature and enhancement nodes and estimates the output weights without deep backpropagation.
・Its standard least-squares training, however, is vulnerable in two different ways: (i) large residuals caused by noise, outliers, or corrupted labels can dominate the objective, and (ii) all samples are treated as equally reliab
cs.LG updates on arXiv.org
Import What You Need: Learning When and How to Augment EHR Graphs with External Knowledge
・arXiv:2609.01839v1 Announce Type: new Abstract: Longitudinal prediction from electronic health records (EHRs) is limited by the sparsity and irregularity in patient trajectories, and knowledge augmentation with external knowledge graphs (KGs) offers a promising way to alleviate these issues.
・However, most existing methods perform fixed, context-agnostic topology augmentation by adding the same KG nodes and edges rega
cs.LG updates on arXiv.org
Improved Gradient Descent Lower Bounds Beyond Nesterov
・arXiv:2609.02855v1 Announce Type: cross Abstract: We study how far gradient descent (GD) can be accelerated by predetermined stepsizes in smooth convex optimization.
・Going beyond the classical $\Omega(n^{-2})$ first-order oracle lower bound of Nemirovsky and Yudin, we prove an $\Omega(n^{-1.6342})$ non-anytime lower bound and an $\Omega(n^{-1.2408})$ anytime lower bound.
・These improve the recent $\Omega(n^{-1.932})$
cs.LG updates on arXiv.org
Improving Evaluation Realism with Inference-Time Compute and Deployment Scaffolds
・arXiv:2609.02302v1 Announce Type: cross Abstract: A core obstacle to alignment evaluation is evaluation awareness: capable models can tell when they are being tested rather than deployed, weakening the conclusions a safety evaluation can support.
・We present two techniques that make simulated alignment evaluations harder to distinguish from real deployments.
・Our first technique, critique refinement, spends additional
Takara TLDR - Daily AI Papers
Improving Evaluation Realism with Inference-Time Compute and Deployment Scaffolds
・A core obstacle to alignment evaluation is evaluation awareness: capable models can tell when they are being tested rather than deployed, weakening the conclusions a safety evaluation can support.
・We present two techniques that make simulated alignment evaluations harder to distinguish from real deployments.
・Our first technique, critique refinement, spends additional inference-time compute on each simulator action: t
cs.LG updates on arXiv.org
Inference-Native Zeroth-Order Optimization
・arXiv:2605.28760v2 Announce Type: replace Abstract: Zeroth-order (ZO) optimization removes backpropagation, but conventional implementations still create candidate states by mutating model weights and materialize updates through the full parameter state.
・We introduce Inference-Native ZO, which exposes ZO's query semantics and lowers candidate-state evaluation and mutable learning state to abstractions an inference ru
cs.LG updates on arXiv.org
Interpretable Symptom Vectors for Depression in a Large Language Model
・arXiv:2609.01832v1 Announce Type: cross Abstract: Patients with depression present with diverse symptom profiles, yet clinical practice routinely reduces this variation to a single severity score.
・Large language models (LLMs) can potentially capture various symptoms and their severity from patient speech.
・However, how depressive symptoms are represented inside LLMs remains poorly understood, limiting clinical trust.
Google DeepMind News
Introducing WeatherNext 3, our most advanced and accurate global weather AI model
Introducing WeatherNext 3, our most advanced and accurate global weather AI model
cs.LG updates on arXiv.org
Language Diffusion Models are Associative Memories Capable of Retrieving Unseen Data
・arXiv:2604.26841v2 Announce Type: replace Abstract: When do language diffusion models memorize their training data, and how to quantitatively assess their true generative regime?
・We address these questions by showing that Uniform-based Discrete Diffusion Models (UDDMs) fundamentally behave as Associative Memories (AMs) $\textit{with emergent creative capabilities}$.
・The core idea of an AM is to reliably recover store
cs.LG updates on arXiv.org
Language Models Can Control Their Own Attention
・arXiv:2609.02737v1 Announce Type: cross Abstract: Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter.
・If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply.
・A prominent approach mitigates this cost by pre-selecting relevan
Takara TLDR - Daily AI Papers
Large Language Models (LLMs) for Telecom Root Cause Analysis (RCA): A Structured Reasoning Framework for Evidence-Grounded Diagnosis
・Root cause analysis (RCA) is a critical task in telecom network operations, but diagnosing performance degradations in modern 5G and emerging 6G networks remains challenging due to complex cross-layer dependencies.
・While large language models (LLMs) offer promising capabilities for reasoning and knowledge integration, directly applying vanilla LLMs to telecom RCA often leads to hallucination, unstable reasoning, and
cs.LG updates on arXiv.org
Latent unified smooth Hamiltonians for excited state chemistry
・arXiv:2609.01871v1 Announce Type: cross Abstract: We describe a neural network architecture and training procedure designed to model electronic ground and excited states of arbitrary molecular systems.
・By indirectly learning a latent, implicit basis representation of the electronic-state Hamiltonian, the model offers a unified treatment of multiple electronic states, conical intersections, and non-adiabatic couplings
cs.LG updates on arXiv.org
Learn from Whoever Is Right: Answer-Verified Multi-Teacher Distillation for Multi-Domain LLMs
・arXiv:2609.02548v1 Announce Type: new Abstract: Modern large language models (LLMs) rely on reinforcement learning to build strong capabilities in individual domains, but integrating those capabilities into a single deployable model remains challenging.
・By routing each sample to the teacher whose domain matches it, existing approaches let a domain label decide which teacher provides supervision.
・However, domain exper
Takara TLDR - Daily AI Papers
Learn from Whoever Is Right: Answer-Verified Multi-Teacher Distillation for Multi-Domain LLMs
・Modern large language models (LLMs) rely on reinforcement learning to build strong capabilities in individual domains, but integrating those capabilities into a single deployable model remains challenging.
・By routing each sample to the teacher whose domain matches it, existing approaches let a domain label decide which teacher provides supervision.
・However, domain expertise holds only on average: the matched teacher
cs.LG updates on arXiv.org
Learning and extrapolating scale-invariant processes
・arXiv:2601.14810v3 Announce Type: replace-cross Abstract: Machine Learning (ML) has deeply changed some fields recently, like Language and Vision and we may expect it to be relevant also to the analysis of of complex systems.
・Here we want to tackle the question of how and to which extent can one regress scale-free processes, i.e.
・processes displaying power law behavior, like earthquakes or avalanches?
cs.LG updates on arXiv.org
Learning Encodings by Maximizing State Distinguishability: Variational Quantum Error Correction
・arXiv:2506.11552v3 Announce Type: replace-cross Abstract: Quantum error correction is crucial for protecting quantum information against decoherence.
・Traditional codes like the surface code require substantial overhead, making them impractical for near-term, early fault-tolerant devices.
・We propose a novel objective function for tailoring error correction codes to specific noise structures by maximizing the distingui
cs.LG updates on arXiv.org
Learning Spectral-Like Mesh-Free Discretisations
・arXiv:2609.02833v1 Announce Type: cross Abstract: Meshfree methods such as smoothed particle hydrodynamics (SPH) with kernel corrections, radial basis function-generated finite differences (RBF-FD), and the local anisotropic basis function method (LABFM) construct discrete differential operators by imposing polynomial consistency on a local stencil.
・For stencils containing more nodes than there are consistency constr
cs.LG updates on arXiv.org
Learning the Constitutive Behavior of Materials via Neural Operators and Causal Attention: Case Studies in Plasticity and Damage
・arXiv:2609.02194v1 Announce Type: new Abstract: Classical constitutive modeling of path-dependent inelastic materials relies on internal state variables whose evolution equations must be postulated based on domain knowledge and calibrated against experimental data.
・However, in many practical settings, the relevant internal variables are typically not measurable in experiments, and the constitutive response must be in
Takara TLDR - Daily AI Papers
Learning the Constitutive Behavior of Materials via Neural Operators and Causal Attention: Case Studies in Plasticity and Damage
・Classical constitutive modeling of path-dependent inelastic materials relies on internal state variables whose evolution equations must be postulated based on domain knowledge and calibrated against experimental data.
・However, in many practical settings, the relevant internal variables are typically not measurable in experiments, and the constitutive response must be inferred entirely from measured strain-stress data
cs.LG updates on arXiv.org
Learning-Based Reconstruction Attacks on Coordinate-Obfuscated Point Clouds
・arXiv:2609.02568v1 Announce Type: cross Abstract: Volumetric video based on point cloud representations enables immersive virtual and augmented reality applications but introduces significant challenges for efficient and secure content delivery.
・Prior work proposed a selective coordinate encryption framework for point clouds that encrypts only a subset of coordinates, reducing computational costs while visually degra
ITmedia NEWS 最新記事一覧
LINE×PayPay連携が延期の見通し 総務省の行政指導受け安全確認に時間 報道
・LINEヤフーが、今夏に予定していた「LINE」と「PayPay」のアカウント連携を延期する方針を固めたと、複数のメディアが9月2日に報じた。新たな開始時期は定まっておらず、利用者情報の保護などの安全確認に時間がかかっているとみられるという。
cs.LG updates on arXiv.org
Linear Fusion MultiDiffusion for Fast Training-Free Spherical Panorama Generation
・arXiv:2609.01997v1 Announce Type: cross Abstract: We propose LF-MultiDiffusion, a training-free panorama generation method that extends MultiDiffusion to support linear projections between target and reference image spaces.
・Our key idea is to reformulate latent aggregation as a regularized least-squares problem and solve it efficiently with a Krylov-based iterative solver inside the denoising loop.
・This formulation e
cs.LG updates on arXiv.org
LLM-as-a-Judge Is Not an Oracle: Why Self-Improving Agents Need Deterministic Guardrails
・arXiv:2609.02246v1 Announce Type: cross Abstract: Self-improving agent pipelines have a problem at their center.
・An optimizer rewrites prompts to score higher, and the score comes from a judge that is itself an LLM.
・That judge has the last word on whether the system is getting better, and our position is that it has not earned it.
cs.LG updates on arXiv.org
Loom: Weaving Diagnostic Strands into Free-Text Consensus via Embedding-Space Reweighting
・arXiv:2609.02649v1 Announce Type: cross Abstract: Aggregating noisy, conflicting textual hypotheses into a reliable consensus is a fundamental challenge when deploying NLP systems in real-world industrial settings.
・While monolithic Large Language Model (LLM) agents offer unbounded expressivity for tasks like Root Cause Analysis (RCA), they suffer from context limits, compounding hallucinations, and prohibitive infere
Takara TLDR - Daily AI Papers
Loom: Weaving Diagnostic Strands into Free-Text Consensus via Embedding-Space Reweighting
・Aggregating noisy, conflicting textual hypotheses into a reliable consensus is a fundamental challenge when deploying NLP systems in real-world industrial settings.
・While monolithic Large Language Model (LLM) agents offer unbounded expressivity for tasks like Root Cause Analysis (RCA), they suffer from context limits, compounding hallucinations, and prohibitive inference latency.
・Traditional weak supervision offers s
cs.LG updates on arXiv.org
LoRA-TSD: Tangent-Space Spectral Descent for LoRA via Muon-Style Updates
・arXiv:2609.02734v1 Announce Type: new Abstract: Low-rank adaptation (LoRA) is the standard way to fine-tune large models, yet when its two factors are trained independently, the update ignores the geometry of the low-rank weight change it induces.
・We introduce LoRA-TSD, an optimizer that treats every LoRA step as a tangent vector of the fixed-rank matrix manifold and takes the spectral-norm steepest-descent step of M
Takara TLDR - Daily AI Papers
LoRA-TSD: Tangent-Space Spectral Descent for LoRA via Muon-Style Updates
・Low-rank adaptation (LoRA) is the standard way to fine-tune large models, yet when its two factors are trained independently, the update ignores the geometry of the low-rank weight change it induces.
・We introduce LoRA-TSD, an optimizer that treats every LoRA step as a tangent vector of the fixed-rank matrix manifold and takes the spectral-norm steepest-descent step of Muon inside that tangent space, mapping the resul
cs.LG updates on arXiv.org
Marginal Expected Revenue for Jointly Ranking Auction and Fixed-Price Listings in E-Commerce Sponsored Search
・arXiv:2609.01628v1 Announce Type: cross Abstract: E-commerce search ranking must balance multiple objectives--relevance, user engagement, and platform revenue--when allocating impression slots to competing listings.
・Estimating the expected revenue component is well understood for fixed-price items, but becomes challenging when marketplace inventory includes mixed listing formats such as pure auctions and hybrid "Auct
cs.LG updates on arXiv.org
MDM-Prime-v2: Binary Encoding and Index Shuffling Enable Scaling of Diffusion Language Models
・arXiv:2603.16077v4 Announce Type: replace Abstract: Masked diffusion models (MDM) exhibit superior generalization when learned using a Partial masking scheme (Prime).
・This approach converts tokens into sub-tokens and models the diffusion process at the sub-token level.
・We identify two limitations of the MDM-Prime framework.
cs.LG updates on arXiv.org
Median-of-Means as an Extremal Convex Estimator and a Nonconvex Route to the Trimmed Oracle
・arXiv:2609.01689v1 Announce Type: new Abstract: We revisit median-of-means estimation from a deterministic optimization viewpoint and develop a family of block-Lp estimators for robust learning with heavy-tailed and adversarially corrupted data.
・In a block contamination model with at least a fraction 1 minus epsilon of good blocks, we first show that every convex block M-estimator has worst-case robustness constant a
cs.LG updates on arXiv.org
MISApp: Multi-Hop Intent-Aware Session Graph Learning for Next App Prediction
・arXiv:2603.21653v2 Announce Type: replace Abstract: Predicting the next mobile app a user will launch is essential for proactive mobile services.
・Yet accurate prediction remains challenging in real-world settings, where user intent can shift rapidly within short sessions and user-specific historical profiles are often sparse or unavailable, especially under cold-start conditions.
・Existing approaches mainly model app
cs.LG updates on arXiv.org
MM++: Post-Hoc Scale-Invariant Multilayer OOD Detection via Top-K Gated Feature Fusion
・arXiv:2606.17352v2 Announce Type: replace Abstract: We introduce MM++ (Multilayer Mahalanobis++), a strictly post-hoc, and scale-invariant framework for Out-of-Distribution (OOD) detection.
・To address the trade-off between scale invariance and hierarchical expressivity, MM++ constructs a principled joint feature space.
・It first identifies discriminative intermediate layers by measuring entropy density drops, which ma
cs.LG updates on arXiv.org
Momentum in large-batch training: Polyak enlarges the critical batch size, Nesterov improves data efficiency
・arXiv:2609.02728v1 Announce Type: cross Abstract: We study when and how momentum improves large-batch training in the one-pass regime, using power-law kernel regression as a tractable setting.
・We first characterize risk stability through the critical learning rate, defined as the largest learning rate for stable training, and obtain $\eta_{\mathrm{SGD}}^{\mathrm{crit}}\eqsim 1$, $\eta_{\mathrm{Polyak}}^{\mathrm{crit}
Takara TLDR - Daily AI Papers
Momentum in large-batch training: Polyak enlarges the critical batch size, Nesterov improves data efficiency
・We study when and how momentum improves large-batch training in the one-pass regime, using power-law kernel regression as a tractable setting.
・We first characterize risk stability through the critical learning rate, defined as the largest learning rate for stable training, and obtain $η_{\mathrm{SGD}}^{\mathrm{crit}}\eqsim 1$, $η_{\mathrm{Polyak}}^{\mathrm{crit}}\eqsim \min\{1,B(1-ρ)\}$, and $η_{\mathrm{Nesterov}}^{\
cs.LG updates on arXiv.org
Monotonic anomaly detection
・arXiv:2410.23158v3 Announce Type: replace Abstract: Semi-supervised anomaly detection is based on the principle that any record that looks different from normal training data is a potential anomaly.
・However, in some cases we are specifically interested in anomalies that correspond to high attribute values (or low, but not both).
・For distance-based methods, we propose an asymmetrical distance measure that takes this m
cs.LG updates on arXiv.org
Morphology signal in whole slide image foundation models can automatically triage slides
・arXiv:2609.01987v1 Announce Type: cross Abstract: Patient exams in the cancer diagnosis and staging process typically generate several whole slide images (WSIs).
・One of the initial steps in training models on WSI data is identifying one or a few slides containing tumor or other diagnostic biomarkers necessary for downstream prediction tasks such as estimating recurrence risk or progression-free survival.
・This step re
cs.LG updates on arXiv.org
Multi-Agent Retrieval-Augmented Generation for Efficient Cloud Knowledge Base Search in Telecom SNOC Environment
・arXiv:2609.01618v1 Announce Type: cross Abstract: Telecom Service and Network Operations Centers (SNOCs) rely on large collections of cloud documents, including Standard Operating Procedures (SOPs), vendor technical manuals, incident reports, and configuration guides, to maintain uninterrupted network operations.
・During critical incidents, engineers must quickly retrieve accurate information, yet traditional keyword
cs.LG updates on arXiv.org
Multi-Mask Diffusion Language Models for Few-Step Generation
・arXiv:2607.19686v3 Announce Type: replace-cross Abstract: Masked diffusion models (MDMs) are a promising family of language generators, but achieving high-quality few-step generation remains challenging.
・In MDMs, all forward trajectories collapse to a single fully masked state, leaving no terminal entropy for consistency-style few-step generation.
・While recent few-step alternatives based on uniform-state diffusion av
stat.ML updates on arXiv.org
Multidimensional scaling of two-mode three-way asymmetric dissimilarities: finding archetypal profiles and clustering
・arXiv:2511.15813v2 Announce Type: replace-cross Abstract: Multidimensional scaling visualizes dissimilarities among objects and reduces data dimensionality.
・While many methods address symmetric proximity data, asymmetric and especially three-way proximity data (capturing relationships across multiple occasions) remain underexplored.
・The h-plot enables the analysis of asymmetric and non-reflexive relationships by embe
cs.LG updates on arXiv.org
Network-Aware Forecasting on Wireless Access Points
・arXiv:2609.01957v1 Announce Type: cross Abstract: Enterprise wireless access points (APs) are promising platforms for predictive machine learning (ML), but their primary responsibility remains providing wireless connectivity and network services.
・Predictive inference must therefore share an AP's CPU and memory with packet processing, Wi-Fi and IoT radio operations, and client management.
・This resource contention crea
Takara TLDR - Daily AI Papers
Network-Aware Forecasting on Wireless Access Points
・Enterprise wireless access points (APs) are promising platforms for predictive machine learning (ML), but their primary responsibility remains providing wireless connectivity and network services.
・Predictive inference must therefore share an AP's CPU and memory with packet processing, Wi-Fi and IoT radio operations, and client management.
・This resource contention creates two risks: a model that performs well on proxy
cs.LG updates on arXiv.org
Neural operators approximate strongly continuous convex monotone semigroups
・arXiv:2609.02727v1 Announce Type: cross Abstract: We approximate strongly continuous convex monotone semigroups by learning their Chernoff-type one-step operators with neural operators.
・First, we introduce the general class of so-called Chernoff-neural operators and show in a universal approximation theorem that they can approximate the Chernoff one-step operators arbitrarily well.
・By using stability estimates betwee
cs.LG updates on arXiv.org
Neural Variational Cut Posteriors without Upstream Data
・arXiv:2510.10268v3 Announce Type: replace-cross Abstract: In many applications, one must propagate parameter uncertainty from an earlier (upstream) analysis, available as samples, to subsequent (downstream) analyses without feedback.
・This problem is called cutting feedback or cut-Bayes, and the cut-posterior, the optimal posterior preserving information-flow constraints, is well characterized.
・However, sampling from
cs.LG updates on arXiv.org
No Data Wasted: A Semi-supervised Generative Model for Incomplete Multi-view Data Integration with Missing Labels
・arXiv:2508.11180v2 Announce Type: replace Abstract: Multi-view learning is widely applied to real-life datasets, but it often suffers from both missing views and missing labels.
・Prior probabilistic approaches addressed the missing view problem by using a product-of-experts scheme to aggregate representations from present views and achieved superior performance over deterministic classifiers, using the information bot
cs.LG updates on arXiv.org
Nonasymptotic CLT and Error Bounds for Linear Two-Time-Scale Stochastic Approximation
・arXiv:2502.09884v4 Announce Type: replace Abstract: We consider linear two-time-scale stochastic approximation algorithms driven by martingale noise.
・Recent applications in machine learning motivate the need to understand finite-time error rates, but conventional stochastic approximation analyses focus on either asymptotic convergence in distribution or finite-time bounds that are far from optimal.
・Prior work on asym
cs.LG updates on arXiv.org
Nova: An End-to-End MLIR Compiler for Deep Learning
・arXiv:2608.00029v3 Announce Type: replace-cross Abstract: The performance of deep learning models at scale relies heavily on how effectively high-level mathematical operations are mapped to underlying physical hardware.
・While high-level tensor frameworks provide flexible abstractions, their execution models inherently lack the whole-graph visibility required to maximize hardware utilization, often forcing a reliance
NVIDIA Blog
NVIDIA to Acquire Hugging Face
・I’m excited to announce that NVIDIA has agreed to acquire Hugging Face for $12,930,300,000.
・Together, we will scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide.
・Over the past decade, Clem, Julien, Thomas and the team at Hugging Face have built something remarkable: a vibrant home for […]
cs.LG updates on arXiv.org
Objective-Behavior Alignment: Diagnostics for MORL Policy Selection
・arXiv:2606.21321v2 Announce Type: replace Abstract: Real-world decision-making often requires optimizing multiple competing objectives simultaneously.
・In reinforcement learning (RL), this is typically addressed by combining reward signals into a single scalar objective via a scalarization function, which can be fragile: small changes in the weights can induce drastically different policies.
・Multi-objective reinforcem
cs.LG updates on arXiv.org
oHC: Orthogonal Hyper-Connections on SO(4) via Quaternions
・arXiv:2609.02672v1 Announce Type: cross Abstract: Hyper-Connections (HC) replace the single residual stream of a Transformer with $n$ parallel ones, mixing them at every layer with a learned $n \times n$ residual matrix.
・Leaving that matrix unconstrained places no limit on the factor by which the mixing step rescales the residual streams, and that factor compounds across layers, which destabilizes training.
Takara TLDR - Daily AI Papers
oHC: Orthogonal Hyper-Connections on SO(4) via Quaternions
・Hyper-Connections (HC) replace the single residual stream of a Transformer with $n$ parallel ones, mixing them at every layer with a learned $n \times n$ residual matrix.
・Leaving that matrix unconstrained places no limit on the factor by which the mixing step rescales the residual streams, and that factor compounds across layers, which destabilizes training.
・Manifold-constrained Hyper-Connections (mHC) address this b
cs.LG updates on arXiv.org
Omega-N: Interpretable Structural Node Descriptors and Their Applicability Domain
・arXiv:2609.01633v1 Announce Type: cross Abstract: A composite structural index summarises a network in one number; for a triangle-based index it is spectrally redundant: Tr(A^3) is the third moment of the adjacency spectrum.
・The non-redundant content sits one level down, in diag(A^3), which depends on eigenvectors and is not spectrally determined.
・A corollary in the theory paper for this index family stated that, and
cs.LG updates on arXiv.org
On Cost-Aware Designs for Sequential Hypothesis Testing
・arXiv:2512.19067v2 Announce Type: replace-cross Abstract: We introduce Cost-Aware (CA) Sequential Hypothesis Testing (CASHT), in which an active decision-maker selects sensing actions with differing, random costs to identify the true hypothesis under an average-error constraint $\delta$ while minimizing the expected total cost rather than the number of samples.
・For fixed costs, we prove that the optimal expected tota
cs.LG updates on arXiv.org
On the Expressive Power and Limitations of Multi-Layer SSMs
・arXiv:2604.14501v2 Announce Type: replace Abstract: We study how depth, finite precision, state dimension, and chain-of-thought (CoT) affect the expressive power of multi-layer state-space models (SSMs).
・For the explicit-table $K$-function-composition problem, a canonical benchmark for sequential information propagation, we prove that any $L$-layer SSM solving $(L+3)$-function composition must satisfy $d^2p=\Omega(N/
cs.LG updates on arXiv.org
On-Policy Distillation Meets Off-Policy GRPO: Training Compact Instruction-Following Rerankers
・arXiv:2609.01947v1 Announce Type: new Abstract: Compact instruction-following rerankers are attractive for deployment, but conventional distillation pipelines typically train students by offline imitation of teacher outputs on a fixed set of examples, constraining supervision to the teacher's observed ranking space.
・We revisit reranker distillation through the lens of reinforcement learning.
・We propose a two-stage fr
cs.LG updates on arXiv.org
One Model, Many Graphs: Learning over Attributed Graphs across Heterogeneous Modalities with Vision-Language Models
・arXiv:2607.19128v2 Announce Type: replace Abstract: Vision-language models (VLMs) provide a unified representation space for textual and visual information, yet their potential as general-purpose backbones for graph-structured data remains largely unexplored.
・In practice, attributed graphs exhibit substantial modality heterogeneity: some graphs contain only textual node attributes, others only visual attributes, whil
stat.ML updates on arXiv.org
Online Multivariate Regularized Distributional Regression for High-dimensional Probabilistic Electricity Price Forecasting
・arXiv:2504.02518v4 Announce Type: replace Abstract: Probabilistic electricity price forecasting (PEPF) is vital for short-term electricity markets, yet the multivariate nature of day-ahead prices - spanning 24 consecutive hours - remains underexplored.
・At the same time, real-time decision-making requires methods that are both accurate and fast.
・We introduce an online algorithm for multivariate distributional regressi
cs.LG updates on arXiv.org
Online Non-Monotone DR-Submodular Maximization Matching the Offline $0.401$ Factor
・arXiv:2609.02145v1 Announce Type: new Abstract: We study online maximization of nonnegative, non-monotone DR-submodular functions over compact convex down-closed subsets of the $d$-dimensional unit cube.
・The best known constructive offline approximation factor is $0.401$ under the corresponding meta-solvability assumptions, whereas comparable adversarial online guarantees had remained at $1/e$.
・We show that this fact
cs.LG updates on arXiv.org
Online Reinforcement Learning in the Met Office Unified Model through Distributed Model-Agent Coupling
・arXiv:2609.02566v1 Announce Type: new Abstract: Machine-learnt corrections can complement numerical weather prediction only if they adapt to the evolving model state while preserving dynamical consistency and numerical stability.
・To test this within a global forecasting model, we couple the Met Office (UKMO) Unified Model (UM) with distributed RL agents through rank-local tensors.
・A DDPG actor shares weights across t
Takara TLDR - Daily AI Papers
Online Reinforcement Learning in the Met Office Unified Model through Distributed Model-Agent Coupling
・Machine-learnt corrections can complement numerical weather prediction only if they adapt to the evolving model state while preserving dynamical consistency and numerical stability.
・To test this within a global forecasting model, we couple the Met Office (UKMO) Unified Model (UM) with distributed RL agents through rank-local tensors.
・A DDPG actor shares weights across the 70 vertical model levels of each atmospheric
cs.LG updates on arXiv.org
Optimal Transport for Network Comparison: A Review with Machine Learning Applications
・arXiv:2608.27500v2 Announce Type: replace-cross Abstract: Network comparison using optimal transport is a growing area of research in network science.
・Unlike standard graph metrics, optimal transport computes both network dissimilarity and a transport plan that explains how one graph morphs into another.
・In this paper, we review how optimal transport compares undirected, unweighted graphs using three primary distance
cs.LG updates on arXiv.org
OptSkills: Learning Generalizable Optimization Skills from Problem Archetypes via Cluster-Based Distillation
・arXiv:2605.29829v2 Announce Type: replace-cross Abstract: Leveraging Large Language Models (LLMs) to automatically formulate and solve optimization problems from natural language has emerged as an efficient paradigm for automated optimization.
・However, existing methods still exhibit limited generalization: they are sensitive to superficial narrative variations, reuse experience mainly at the case level, and struggle
cs.LG updates on arXiv.org
OR-Transformer: Scaling Real-Time Decision-Making to 1,000 Items
・arXiv:2609.01933v1 Announce Type: new Abstract: Modern supply chain operations can require coordinating replenishment across thousands of heterogeneous items under correlated stochastic demand, heterogeneous lead times, and shared fixed ordering costs, yielding observation spaces exceeding $10^4$ dimensions.
・At this scale, rolling-horizon stochastic mixed-integer linear programs (MILPs) become prohibitively slow, whi
cs.LG updates on arXiv.org
Oracle, will I ever learn? A study of prediction convergence and complementarity across link prediction models
・arXiv:2609.02638v1 Announce Type: new Abstract: Knowledge graphs have become an important source of structured knowledge for Web applications, including search, question answering, and recommender systems.
・In these applications, link prediction can serve either as a prediction task itself or as a means to enrich incomplete knowledge graphs for downstream tasks.
・Interestingly, different link prediction models, or even
cs.LG updates on arXiv.org
Orthogonal Ensembles and Tested Explanations for Performer-Independent Body-Motion Emotion Recognition
・arXiv:2609.02510v1 Announce Type: cross Abstract: We study body-only, 12-class acted-emotion classification from skeleton motion under leave-performer-out (LPO) evaluation, a hard, underdetermined setting: chance is 8.3%, and a protocol-matched reproduced STGCN++ baseline reaches only 25.73 +/- 4.03% Macro-F1.
・We show that reliable gains come not from a new architecture but from combining eleven models with orthogona
cs.LG updates on arXiv.org
OutageDiT: A Generative Foundation Model for Power Outage Forecasting and Scenario Simulation
・arXiv:2609.01896v1 Announce Type: new Abstract: Power-outage planning requires scenarios before an event occurs.
・These scenarios must represent uncertainty in magnitude, timing, and duration while preserving temporal dependence.
・However, severe events are rare, and data from any single region contain few examples of extreme outage and restoration patterns.
Takara TLDR - Daily AI Papers
PaperCompiler: Faithful Paper-to-Code Generation via Repository-Level Specification Compilation
・Faithfully translating research papers into repository-level implementations remains challenging because papers often describe methods at a high level, leave implementation assumptions implicit, and require generated repositories to preserve method logic, evaluation protocols, and cross-file consistency.
・Despite recent advances in paper-to-code agents, their intermediate outputs are often presented as free-form plans
cs.LG updates on arXiv.org
Perceptually Regularized Diffusion Model for Image Super-Resolution
・arXiv:2609.02016v1 Announce Type: cross Abstract: Image super-resolution, which aims to reconstruct high-resolution images from their low-resolution observations, is fundamental to medical imaging, remote sensing, surveillance, microscopy, and scientific visualization.
・Traditional model-based methods formulate super-resolution as an inverse problem with hand-crafted regularization priors.
・While interpretable and theo
cs.LG updates on arXiv.org
Percolation Dynamics in Optimization : Variance Cascades and Discrete Scale Invariance
・arXiv:2609.02373v1 Announce Type: new Abstract: We study the dynamics of Stochastic Gradient Descent (SGD), which is known to steer deep neural networks toward invariant sets that correspond to simpler subnetworks.
・How this steering unfolds over time remains poorly understood.
・We answer this by modeling the stochastic gradient flow (SGF) as a percolation process, in which architectural symmetries force subnetworks to
cs.LG updates on arXiv.org
Persistent Sparse Autoencoders: Learning Feature-Specific Timescales in Language Model Representations
・arXiv:2607.17117v2 Announce Type: replace Abstract: Sparse autoencoders (SAEs) decompose language model activations into sparse features, yet these models traditionally encode each token independently, failing to expose information that persists across a sequence.
・We first show that temporal persistence can naturally emerge in standard SAE features: after a feature activates, the hidden state remains aligned with its
Takara TLDR - Daily AI Papers
PGPO: Potential-Guided Policy Optimization for Multi-Turn Agentic Tasks
・Group-based reinforcement learning (RL) has become an effective paradigm for LLM post-training, but in multi-turn agentic tasks with sparse terminal rewards, it often provides coarse credit for intermediate actions.
・To obtain more fine-grained credit assignment, recent work such as GiGPO introduces step-level advantages for intermediate actions.
・However, these step-level signals still rely on the final outcome of eac
cs.LG updates on arXiv.org
Poisoning Attacks on the PGM-index
・arXiv:2609.02328v1 Announce Type: cross Abstract: The PGM-index (Ferragina and Vinciguerra, VLDB'20) is one of the most practical learned indexes, owing to its theoretical elegance and consistently strong empirical performance.
・It is built on optimal piecewise linear approximations (PLAs) that minimize the number of segments.
・In this paper, we ask how sensitive this optimal PLA itself is to poisoning attacks.
cs.LG updates on arXiv.org
Pooling and Drift in Delayed Bandits
・arXiv:2609.01761v1 Announce Type: cross Abstract: A system often has to act long before it learns whether the act worked: a recommender sees a click in seconds and a purchase in days.
・With $K$ actions and a delay of $d$ rounds, the best rate known for this setting is $\widetilde{O}(\sqrt{(K+d)T})$ over $T$ rounds, so a longer menu is always more expensive to learn from.
・It need not be: if the outcome depends on the a
cs.LG updates on arXiv.org
Post-Training Language Models for Gold-Medal Performance in Coding Competitions
・arXiv:2609.02849v1 Announce Type: new Abstract: Competitive programming has become a key test of large language model reasoning, with international competitions such as IOI and ICPC representing its most challenging settings.
・We present an end-to-end specialization pipeline combining large-scale problem curation, synthetic reasoning traces, supervised fine-tuning (SFT), and reinforcement learning (RL).
・Using 22,000 c
cs.LG updates on arXiv.org
Post-Training Ternarization of Qwen3-4B Capability, Effective Bit Budget, Storage Compression, and Deployment
・arXiv:2609.01962v1 Announce Type: cross Abstract: Ultra-low-bit language models can reduce storage and memory bandwidth, but a nominal "1.58-bit" label does not fully describe the stored representation, retained capability, or runtime behavior.
・We study an end-to-end post-training conversion of Qwen, an instruction-tuned 4B-parameter model, using KOTMS rotation, E2M-ATQ ternarization, and GPTQ-style error compensatio
cs.LG updates on arXiv.org
Posterior Tempering Explains Variance Inflation in Linear and Generalized Linear Thompson Sampling
・arXiv:2609.01999v1 Announce Type: cross Abstract: We study a variant of the Thompson Sampling (TS) algorithm, called $\alpha$-TS, for solving stochastic generalized linear bandit problems.
・Existing analyses of TS require inflating the posterior variance to derive near-optimal regret guarantees.
・We formalize the idea of variance inflation by introducing $\alpha$-TS that uses a fractional or $\alpha$-posterior instead
＠IT 全フォーラム 最新記事一覧
PR： 「見えないOT資産、見えないセキュリティリスク」を解消する IT/OTの分断を解消する方法とは
・実態が把握できず、本社IT部門／情報セキュリティ部門が対策にも十分に介入できないことがOTセキュリティの難しいところだ。工場の稼働に影響を与えずに十分な対策をとる方法を探る。
＠IT 全フォーラム 最新記事一覧
PR： AIを開発組織全体に展開するには 日本IBMが示す「3つの壁」とAI駆動開発の実践法
・AIの企業利用が広がる一方、「構造」「基盤」「人」という壁によって、全社展開やレガシーモダナイゼーションに悩む企業は多い。複雑な大規模開発において、AIを真の「組織の武器」に変えるにはどうすべきか。日本IBMが提唱するAI駆動開発の最新アプローチと実践の鍵を解き明かす。
＠IT 全フォーラム 最新記事一覧
PR： 暴走するAIエージェントにどう気づくか 挙動の異常を捉える「ABA」という新しい行動分析
・生成AIの活用が一問一答の「対話」からタスクを自律処理する「AIエージェント」へ移行しているが、ガードレールを過信するのは危険だ。AIエージェントの暴走や乗っ取りに気づき、新たな内部脅威から自社を守るための対策とは。
cs.LG updates on arXiv.org
PRISM: An Agentic Multi-Model Architecture for Proactive Safety in Autonomous Transportation Systems
・arXiv:2609.01623v1 Announce Type: cross Abstract: Autonomous and intelligent transportation systems operate in complex urban environments where safety depends on interactions among vehicle behavior, environmental conditions, and vulnerable road users (VRUs) such as pedestrians and cyclists.
・Most advanced driver assistance systems (ADAS) employ reactive mechanisms that activate only after hazards have emerged, a criti
Takara TLDR - Daily AI Papers
Privacy Washing: Detecting Internal Contradictions in Privacy Policies
・Privacy policies may contain internal contradictions in which commitments are undermined by practices documented elsewhere in the same policy.
・We operationalize this phenomenon, privacy washing, through a four-stage pipeline: statement extraction, compatibility filtering and natural language inference screening, multi-model judge verification, and thematic analysis, with contradictions confirmed by majority vote of a
cs.LG updates on arXiv.org
Private Computation Space: Experience with Trusted Multi-Cluster Federated Learning for Agriculture
・arXiv:2609.01667v1 Announce Type: cross Abstract: Artificial Intelligence has shown to help improve agricultural practices, yet adoption remains limited: 69% of U.S.
・farmers have privacy concerns with sharing their data, and these concerns must be addressed before adoption is widespread.
・While Federated Learning has been demonstrated to protect privacy at scale for other sectors, deploying a system for agriculture co
cs.LG updates on arXiv.org
ProbeMatchDTI: Probe-Driven Multi-Scale Biochemical Pattern Matching for Drug-Target Interaction Prediction
・arXiv:2609.02549v1 Announce Type: new Abstract: Drug-target interaction (DTI) prediction is an important task in AI-driven drug discovery.
・Although recent biochemical representation learning methods have improved DTI prediction, their passive feature aggregation tends to favor dominant molecular patterns while suppressing weak yet binding-relevant signals, such as functional groups and residue-context patterns, limit
cs.LG updates on arXiv.org
Probing Cultural Signals in Large Language Models through Author Profiling
・arXiv:2603.16749v3 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly deployed in applications with societal impact, raising concerns about the cultural biases they encode.
・We probe these representations by evaluating whether LLMs can perform author profiling from song lyrics in a zero-shot setting, inferring singers' gender and ethnicity without task-specific fine-tuning.
Takara TLDR - Daily AI Papers
Progressive Pseudo-Label Optimization for Point-Supervised Change Detection
・Point-supervised change detection (PS-CD) aims to identify pixel-level changes between bi-temporal images using only sparsely annotated points.
・Although point annotations substantially reduce labeling costs, their limited spatial coverage often results in incomplete and noisy pseudo-labels.
・To address this issue, we propose a two-stage framework that introduces SAM2 priors into PS-CD and progressively adapts them to
cs.LG updates on arXiv.org
Prompt-Space Meta-Learning Does Not Transfer Across Users: A Frozen-LLM Negative Result
・arXiv:2609.01615v1 Announce Type: new Abstract: Personalizing a frozen large language model (LLM) to individual users is often framed as a meta-learning problem in prompt space: each user is a task, and one seeks a shared natural-language adaptation policy that, given a handful of the user's labeled interactions, configures the frozen model for that user.
・The framing is attractive because it is backbone-agnostic and
cs.LG updates on arXiv.org
Prompting the Unknown: Understanding Response Uncertainty in Large Language Models
・arXiv:2407.14845v4 Announce Type: replace Abstract: Large language models (LLMs) are widely used in decision-making across diverse domains.
・Ensuring the generation of safe and reliable responses is critical for the effective deployment of LLM-based applications, particularly in high-stakes domains such as healthcare and finance.
・Most of these applications typically use carefully crafted prompts to guide response gene
cs.LG updates on arXiv.org
Prototype-guided transfer of sparse literature knowledge for electrolyte additive discovery
・arXiv:2609.02209v1 Announce Type: cross Abstract: Electrolyte additive discovery remains challenging because experimentally validated molecules are sparse, whereas accessible chemical spaces are vast and largely unlabeled.
・This challenge is amplified in lithium-ion batteries, where additive performance arises from coupled interfacial reactions rather than a single molecular property.
・Here, we develop a prototype-guid
cs.LG updates on arXiv.org
Pushing Forward Multi-Secret-Key Homomorphic Encryption for Private Average Aggregation
・arXiv:2609.01945v1 Announce Type: cross Abstract: Federated Learning enables multiple clients to train a shared model while keeping their local datasets isolated.
・However, the exchanged model updates may still leak sensitive information, making private aggregation a central building block in practical deployments, especially in the cross-silo setting.
・Homomorphic Encryption naturally fits the client--aggregator commu
cs.LG updates on arXiv.org
Quantum Maximum Likelihood Prediction via Hilbert Space Embeddings
・arXiv:2602.18364v4 Announce Type: replace-cross Abstract: Maximum likelihood prediction (MLP) is a core task at the heart of modern large language models.
・Here, we study a quantum version of this task for a simplified data model consisting of independent and identically distributed samples, as a first step.
・The quantum maximum likelihood predictor (QMLP) is obtained by embedding of empirical probability distributions
cs.LG updates on arXiv.org
Quantum MeanFlow: single-shot generative sampling on NISQ hardware
・arXiv:2609.02186v1 Announce Type: cross Abstract: Quantum generative models offer a promising framework for exploring whether quantum computation can enhance generative machine learning.
・Flow matching is a generative method in which samples are generated by transporting a simple, known distribution to the target data distribution with a learned velocity field.
・Its quantum counterpart, known as quantum flow matching (
cs.LG updates on arXiv.org
Quantum Speedups for Sampling and Non-convex Optimization with Stochastic Oracles
・arXiv:2504.03626v2 Announce Type: replace-cross Abstract: We present quantum speedups for sampling from distributions of the form $\pi\propto e^{-f}$ on $\mathbb{R}^d$.
・We consider two stochastic oracle models: a stochastic gradient oracle, where $f=\frac{1}{n}\sum_{i=1}^n f_i $ and component gradients $\{\nabla f_i\}_{i \in [n]}$ are available, and a stochastic evaluation oracle, where only noisy values of $f$ are a
cs.LG updates on arXiv.org
Random Forest-Informed Cellular Automaton for Large-Scale Wildfire Spread Modelling
・arXiv:2609.01675v1 Announce Type: cross Abstract: Accurate large-scale wildfire spread modelling requires models that capture both the environmental conditions associated with fire occurrence and the local dynamics of fire propagation.
・We propose a three-stage framework that combines a Random Forest (RF) model with a cellular automaton (CA).
・First, an RF model trained on the 2021 Canadian fire season estimates daily
cs.LG updates on arXiv.org
RCProb: Probabilistic rule extraction from classification tree ensembles
・arXiv:2604.25304v2 Announce Type: replace Abstract: Tree ensembles provide strong classification performance but usually behave as black-box models.
・Post-hoc interpretability techniques such as RuleCOSI+ extract a small ruleset that approximates the ensemble, but this simplification can leave the probabilities attached to the extracted rules unreliable.
・In particular, RuleCOSI+ assigns empirical class probabilities t
cs.LG updates on arXiv.org
RecEvolve: A Knowledge-Driven Autonomous Agent System for Recommender Systems
・arXiv:2609.01622v1 Announce Type: cross Abstract: The rise of agentic AI has catalyzed a shift toward self-iterating systems, opening new frontiers for the autonomous optimization of production recommender models.
・This paper presents the empirical validation of a knowledge-driven autonomous agent system, deployed directly on a production large-scale Two-Tower retrieval model.
・By delegating the entire research lifecyc
cs.LG updates on arXiv.org
RecKAN: Kolmogorov-Arnold Networks with a Learnable Recursive Polynomial Basis
・arXiv:2609.01729v1 Announce Type: new Abstract: Kolmogorov--Arnold Networks (KANs) replace the fixed scalar weights of a standard network with learnable univariate functions on each edge, but existing variants still fix the \emph{basis} that those functions are built from: B-splines, Chebyshev polynomials, wavelets, or Jacobi polynomials, and learn only the combination weights over it.
・We introduce RecKAN, which inst
cs.LG updates on arXiv.org
Recursive Value Learning for Long-Horizon Offline Goal-Conditioned RL
・arXiv:2609.02237v1 Announce Type: new Abstract: Scaling offline goal-conditioned reinforcement learning (GCRL) to long-horizon tasks is difficult because (1) long-range value learning depends on shorter-range estimates that may still be inaccurate, and (2) max-based value backups can amplify overestimation through repeated propagation.
・We propose DCRL (Divide-and-Conquer RL), which recursively decomposes each traject
cs.LG updates on arXiv.org
Reinforcement Learning and Rule-Based Peer-to-Peer Pricing in Residential PV-BES Communities
・arXiv:2609.01680v1 Announce Type: new Abstract: This paper compares rule-based and learning-based pricing mechanisms for peer-to-peer (P2P) electricity trading in residential photovoltaic communities.
・The rule-based benchmarks comprise bill-sharing as an ex post allocation mechanism, the mid-market rate, and supply-demand-ratio pricing.
・The reinforcement-learning (RL) formulation is implemented through a Deep Q-Netwo
cs.LG updates on arXiv.org
Reinforcement Learning for Heterogeneous Sensor Selection in Maritime Surveillance
・arXiv:2607.22667v2 Announce Type: replace-cross Abstract: This paper presents an information-gain-guided reinforcement-learning sensor-selection framework for single-vessel tracking in heterogeneous maritime sensor networks.
・The proposed approach is motivated by information-theoretic sensor management: instead of activating all sensors or repeatedly performing computationally expensive online expected-information-gai
cs.LG updates on arXiv.org
Reinforcement learning to choose optimizers
・arXiv:2609.01811v1 Announce Type: cross Abstract: No single optimization method is uniformly best for all problems, and the most suitable optimizer choice can change during a run.
・Existing approaches that change optimizer during execution typically predetermine part of the strategy: the portfolio is restricted to one algorithm class, the switch occurs once at a fixed time, or the frequency of decisions is treated as
cs.LG updates on arXiv.org
Rethinking the Teacher-Student Framework for Test-Time Adaptation
・arXiv:2609.02507v1 Announce Type: new Abstract: Test-Time Adaptation (TTA) has recently emerged as a promising strategy that allows the adaptation of pre-trained models to changing data distributions at deployment time, without access to any labels.
・To mitigate error accumulation, researchers have widely adopted the teacher-student framework, though its long-term stability is often taken for granted.
・In this work, we
Takara TLDR - Daily AI Papers
Retrosynthesis of Synthetic Media for Explainable AI Provenance Forensics
・With the rapid proliferation of generative models on Machine Learning as a Service (MLaaS) platforms, reliably tracing the provenance of synthetic media without modifying generator architectures or parameters remains a major challenge.
・In this work, we propose a self-referential retrosynthesis framework for explainable AI provenance forensics under a fixed-generator setting.
・The framework leverages a jointly optimize
cs.LG updates on arXiv.org
RideSkill: A Hierarchical Algorithm for Generalized Ride Sharing with LLM-Driven Automatic Evolution
・arXiv:2609.02250v1 Announce Type: cross Abstract: Ride-sharing, which allows multiple passengers with different origin-destination (OD) pairs to share a single vehicle, is a challenging operational problem, as it requires orders with different OD pairs to be efficiently bundled and assigned to vehicles under uncertain and varying scenarios.
・Although multi-agent reinforcement learning (MARL) solutions have achieved pr
cs.LG updates on arXiv.org
RINSE: Robust Target-Time Normality Estimation for Zero-Shot Graph Anomaly Detection
・arXiv:2609.02497v1 Announce Type: new Abstract: Zero-shot graph anomaly detection seeks to deploy a detector trained on source graphs to unseen, unlabeled targets, yet domain shift can make source-derived notions of normality unreliable.
・We introduce RINSE (Robust Iterative Normality Self-Estimation), a gradient-free target-time framework that keeps the source-trained detector fixed while sequentially estimating targ
stat.ML updates on arXiv.org
Robust Bayesian Inference for Unnormalized Models with Mixed-Domain Data
・arXiv:2609.01783v1 Announce Type: cross Abstract: Many statistical models involve parameter-dependent normalizing constants that are computationally intractable, creating substantial obstacles to standard Bayesian inference.
・Although existing likelihood-based algorithms can often circumvent these constants, their uncertainty quantification may be poorly calibrated under model misspecification.
・To address these challe
cs.LG updates on arXiv.org
Robust Streaming PCA
・arXiv:1902.03223v4 Announce Type: replace-cross Abstract: We consider streaming principal component analysis when the stochastic data generating model is subject to perturbations.
・While existing models assume a fixed covariance, we adopt a robust perspective where the covariance matrix belongs to a temporal uncertainty set.
・Under this setting, we provide fundamental limits on convergence of any algorithm recovering p
cs.LG updates on arXiv.org
SABER-Math: Automated Benchmark for Information Retrieval Evaluation in Mathematics
・arXiv:2606.29894v2 Announce Type: replace-cross Abstract: As agentic AI systems tackle more complex mathematical tasks, they increasingly rely on information retrieval (IR) to search problem databases, theorem libraries, and educational resources.
・However, choosing the right retriever remains difficult, as it is infeasible to directly isolate its effect on downstream performance.
・On the other hand, existing retrieval
Takara TLDR - Daily AI Papers
SafeEvolve: Harness-Policy Co-Evolution from Agent Experience for Safety Alignment
・The performance of LLM-based agents is jointly shaped by the base model and the harness used when interacting with the environment.
・This exposes them to safety risks in both harmful final responses and multi-step execution trajectories.
・Existing safety alignment mechanisms often rely on either external harness updates or policy optimization, yet applying either paradigm in isolation fails to bridge runtime control wi
cs.LG updates on arXiv.org
Sample Complexity of Linear Quadratic Regulator Without Initial Stability
・arXiv:2502.14210v4 Announce Type: replace-cross Abstract: Inspired by REINFORCE, we introduce a novel receding-horizon algorithm for the Linear Quadratic Regulator (LQR) problem with unknown dynamics.
・Unlike prior methods, our algorithm avoids reliance on two-point gradient estimates while maintaining the same order of sample complexity.
・Furthermore, it eliminates the restrictive requirement of starting with a stable
cs.LG updates on arXiv.org
Scalable Bayesian Optimization of Composite Functions for Image-Based Inverse Problems in Materials Characterization
・arXiv:2609.02126v1 Announce Type: new Abstract: Estimating physical parameters from scientific images is a common inverse problem in materials characterization that often relies on expensive physics-based simulations.
・In electron microscopy, specimen thickness and crystal mistilt are critical parameters that govern how electrons scatter through the sample, and therefore the accuracy of any atomic-scale structure reco
cs.LG updates on arXiv.org
Scalable Direction-Following TTS via Voice Impression-Guided Pseudo Triplet Construction
・arXiv:2609.02623v1 Announce Type: cross Abstract: Voice actors often re-read the same script while modifying their delivery in response to performance directions.
・We study this setting as direction-following TTS, where a system generates a new utterance that reflects a given direction relative to a reference utterance while preserving speaker identity and linguistic content.
・A key challenge is the lack of training da
Takara TLDR - Daily AI Papers
Scalable Direction-Following TTS via Voice Impression-Guided Pseudo Triplet Construction
・Voice actors often re-read the same script while modifying their delivery in response to performance directions.
・We study this setting as direction-following TTS, where a system generates a new utterance that reflects a given direction relative to a reference utterance while preserving speaker identity and linguistic content.
・A key challenge is the lack of training data capturing such relative modifications.
cs.LG updates on arXiv.org
Scalable Kronecker-Fisher Approximation: Efficient Hessian Analysis for Billion-Parameter Language Models Compression
・arXiv:2609.02451v1 Announce Type: new Abstract: In this paper, we propose a scalable Kronecker-based approximation that captures cross-layer interactions without storing the entire Fisher matrix, enabling practical Hessian analysis for billion-parameter networks where full computation is infeasible.
・Our approach reveals consistent vulnerability patterns: value projection layers exhibit the highest sensitivity and str
cs.LG updates on arXiv.org
Schr\"odinger Bridges on Lie Group Manifolds for Probabilistic Intrinsic Generation
・arXiv:2609.02196v1 Announce Type: cross Abstract: Generative modeling directly on geometric manifolds can avoid errors introduced by flattening non-Euclidean data, repeated ambient projection, and coordinate inconsistency in Euclidean representations.
・Schrodinger bridges provide a probabilistic generative framework for entropy-regularized transport between prescribed endpoint distributions.
・We study Schrodinger bridg
cs.LG updates on arXiv.org
SEAL: Reinforcing Global Safety in Mixture-of-Experts through Shared Expert ALignment
・arXiv:2609.02293v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) is a scaling architecture for large language models that activates only a small subset of expert modules per token, enabling massive parameter growth with nearly constant computation.
・Recent Hybrid MoE architecture adds \textit{shared experts} to capture consistently useful representations, further improving stability and generalization.
cs.LG updates on arXiv.org
Secure AI-Driven Super-Resolution for Real-Time Mixed Reality Applications
・arXiv:2512.15823v3 Announce Type: replace-cross Abstract: Immersive formats such as 360{\deg} and 6DoF point cloud videos require high bandwidth and low latency, posing challenges for real-time AR/VR streaming.
・This work focuses on reducing bandwidth consumption and encryption/decryption delay, two key contributors to overall latency.
・We design a system that downsamples point cloud content at the origin server and ap
Takara TLDR - Daily AI Papers
Seeing Beyond the Lesion: Disease Recognition from Reactive CNS Tissue
・Sampling error yields exclusively reactive, non-lesional brain parenchyma in a significant proportion of intracranial biopsies, leaving the underlying disease undiagnosed.
・We benchmark four pathology foundation models (UNI2-h, Virchow2, Prov-GigaPath, H-optimus-0) as frozen patch encoders within a shared attention-based multiple-instance learning framework using 245 whole-slide images from 186 patients with confirmed
cs.LG updates on arXiv.org
Shiva-DiT: Residual-Based Differentiable Top-$k$ Selection for Efficient Diffusion Transformers
・arXiv:2602.05605v2 Announce Type: replace Abstract: Diffusion Transformers (DiTs) are costly at high resolution because self-attention scales quadratically with token sequence length.
・Existing pruning methods do not jointly provide end-to-end learnability, low training overhead, and deterministic token counts for predictable token-dependent computation.
・We propose Shiva-DiT, based on Residual-Based Differentiable Top
cs.LG updates on arXiv.org
Shortcomings and capacities of real-constrained neural networks in complex spaces
・arXiv:2606.04390v3 Announce Type: replace Abstract: We find the asymptotic ratio between the storage capacities when enforcing real pre-activations in a complex hypothesis class as opposed to complex ones in the same class.
・We use weights drawn from the complex Gaussian, which converge asymptotically in norm to the square root of dimension almost surely.
・Our methods depend on Gardner volume-type comparisons at critic
cs.LG updates on arXiv.org
Sim2Signal: Sim-to-Real Benchmarks for Traffic Signal Control
・arXiv:2609.01676v1 Announce Type: new Abstract: Reinforcement learning achieves strong traffic signal control performance in simulation, yet policies trained in simulators often fail once deployed in the real world, a failure known as the Sim-to-Real gap.
・When RL is applied to traffic signal control, this gap arises from several sources: sensing, action execution, traffic dynamics, and the control objective.
cs.LG updates on arXiv.org
Similarity-Aware Personalized Federated Learning in Heterogeneous Environments
・arXiv:2609.02241v1 Announce Type: new Abstract: Federated Learning (FL) allows decentralized clients to train models collaboratively while preserving data privacy.
・However, distribution mismatch across clients often leads to poor global generalization and degraded local client-level performance.
・In such scenarios, some of the clients with their local models trained solely on local data may perform better than the glo
cs.LG updates on arXiv.org
Simulating Classification Models for Ex-Ante Evaluation of Predict-Then-Optimize Methods
・arXiv:2509.02191v3 Announce Type: replace Abstract: Predict-Then-Optimize combines machine learning predictions with downstream optimization to support decision-making when problem parameters are unknown at the time of solving.
・However, better predictive performance does not necessarily lead to better decisions, making it useful to assess this relationship before investing in the development of a prediction model.
cs.LG updates on arXiv.org
SMart: A Multi-source Multi-phase Time Series Representation Transfer Framework
・arXiv:2609.02203v1 Announce Type: new Abstract: Time series representation learning (TSRL) has attracted growing research interests in recent years.
・Two recent explorations in TSRL are: i) exploiting a transformer-based framework to learn time series; ii) instead of using only the targeted dataset, borrowing time series from other datasets to to facilitate representation transfer.
・While these two explorations are sho
cs.LG updates on arXiv.org
Smoothed Analysis for Learning Concepts with Low Intrinsic Dimension
・arXiv:2407.00966v3 Announce Type: replace Abstract: In traditional models of supervised learning, the goal of a learner-- given examples from an arbitrary joint distribution on $\mathbb{R}^d \times \{\pm 1\}$-- is to output a hypothesis that is competitive (to within $\epsilon$) of the best fitting concept from some class.
・In order to escape strong hardness results for learning even simple concept classes, we introdu
cs.LG updates on arXiv.org
SocialBuddy: Tailoring Search Agent for Social Scenarios
・arXiv:2609.01641v1 Announce Type: cross Abstract: In the era of digital social interaction, searching friends' posts from massive social streams has become a fundamental user need.
・However, while modern agentic search frameworks have achieved remarkable success in conventional retrieval tasks, they break down when confronted with heterogeneous user queries and multi-dimensional social feeds, resulting in severe perfo
cs.LG updates on arXiv.org
SoK: Where Do Flow Labels Come From? Auditing Label Provenance in Encrypted Traffic Benchmarks
・arXiv:2609.02140v1 Announce Type: cross Abstract: Encrypted traffic classification infers semantics beyond the flow record from transport-layer observables, and supervised training rests on labels that hold for the individual flow they are attached to.
・Recent systematizations scrutinize model in- puts and data splits; we systematize the complementary label side.
・Across 14 audited benchmark entries, we identify two re
cs.LG updates on arXiv.org
Source Distribution Estimation by Posterior Averaging
・arXiv:2609.02622v1 Announce Type: new Abstract: Simulation-based science often requires a distribution over simulator parameters whose push-forward reproduces a set of real observations: this is the source distribution estimation (SDE) problem.
・Existing methods fit the source against a likelihood surrogate trained once from a fixed proposal prior.
・Their objective is therefore stated only in terms of the surrogate ins
cs.LG updates on arXiv.org
Source-Free Class Relearning: Diagnosing Forgetting in Class Unlearning
・arXiv:2609.02018v1 Announce Type: new Abstract: Class unlearning aims to remove a model's ability to recognize designated forget classes while preserving performance on retain classes.
・However, low forget accuracy after unlearning does not necessarily mean the class structure has been erased.
・Approximate unlearning methods can alter classifier decision boundaries while leaving recoverable structure in the representat
cs.LG updates on arXiv.org
SPADE: SPaT Attack Detection from the Connected Vehicle's Perspective
・arXiv:2609.02741v1 Announce Type: cross Abstract: Signal Phase and Timing (SPaT) messages are a cornerstone of connected vehicle (CV) safety, enabling CVs to perceive and respond to intersection state through Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) communication.
・The integrity of these messages is threatened by a range of application-layer attacks that can bypass conventional authentication when
Takara TLDR - Daily AI Papers
SPADE: SPaT Attack Detection from the Connected Vehicle's Perspective
・Signal Phase and Timing (SPaT) messages are a cornerstone of connected vehicle (CV) safety, enabling CVs to perceive and respond to intersection state through Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) communication.
・The integrity of these messages is threatened by a range of application-layer attacks that can bypass conventional authentication when a roadside unit or peer vehicle is compromised.
NVIDIA Blog
Sparks Fly: NVIDIA Accelerates Local AI at IFA 2026
・Frontier intelligence is going local.
・At IFA 2026, NVIDIA, Microsoft and its partners are teaming up to provide faster inference and new tools that make agents easier to set up and run locally on NVIDIA hardware.
・New compact NVIDIA RTX Spark Windows PCs are also coming in October to give AI enthusiasts, developers and creators […]
cs.LG updates on arXiv.org
Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens
・arXiv:2609.01936v1 Announce Type: cross Abstract: A language model's prediction of its next token develops across layers, and lens methods track this process by decoding intermediate hidden states into tokens.
・But a lens reading reflects both the hidden state and the readout (the unembedding matrix) used to decode it.
・Many lenses are fit on a corpus, and we show that two lenses differing only in their fitting corpus
cs.LG updates on arXiv.org
Spectral Initialization and Scheduled Graph Smoothness for Uncertain Knowledge Graph Completion
・arXiv:2609.02519v1 Announce Type: new Abstract: Uncertain knowledge graphs (UKGs) extend knowledge graphs by assigning each triple a continuous confidence score.
・Since most possible triples lack observed confidences, recent methods rely on semi-supervised learning to generate pseudo-labels.
・These methods initialize entity embeddings without using the confidence-weighted graph, discarding its global community and hub
cs.LG updates on arXiv.org
SpecXMaster Technical Report
・arXiv:2603.23101v4 Announce Type: replace Abstract: Intelligent spectroscopy serves as a pivotal element in AI-driven closed-loop scientific discovery, functioning as the critical bridge between matter structure and artificial intelligence.
・However, conventional expert-dependent spectral interpretation encounters substantial hurdles, including susceptibility to human bias and error, dependence on limited specialized
cs.LG updates on arXiv.org
Stabilizing Private LASSO under Heterogeneous Covariates via Anisotropic Objective Perturbation
・arXiv:2605.01492v2 Announce Type: replace-cross Abstract: We study high-dimensional LASSO under differential privacy via objective perturbation with heterogeneous covariate scales.
・In practical scenarios, covariates often exhibit diverse scales; however, standard preprocessing is problematic under privacy constraints, as it consumes additional privacy budget.
・This heterogeneity induces effective anisotropy in the obj
cs.LG updates on arXiv.org
Stream-CQSA: Exact Out-of-Memory Recovery for Attention
・arXiv:2604.20819v2 Announce Type: replace Abstract: Long-context large language models are limited not only by attention cost but also by out-of-memory (OOM) failures.
・A selected attention call may not fit in available device memory even when the kernel is optimized.
・Exact and approximate attention methods reduce memory use, but every fixed implementation still has a device-specific capacity boundary.
Takara TLDR - Daily AI Papers
Subcellularly Resolved Single-Cell Embedding Learning with Transcriptomic data, Protein Structure and Localization Information
・Existing cell embedding methods predominantly rely on transcriptomic or proteomic measurements and represent each cell as a holistic entity, thereby overlooking the subcellular localization of individual molecules.
・Moreover, they rarely incorporate protein structural information, despite its fundamental role in determining molecular interactions and functions.
・In this work, we propose a multimodal framework for learn
Takara TLDR - Daily AI Papers
TAME: Temporal-Aware Mixture-of-Experts for Text-Video Retrieval
・Text-Video Retrieval (TVR) retrieves videos that match a natural-language query, but extending image-text models such as CLIP to videos is fundamentally limited by the lack of temporal modeling.
・Videos exhibit frame-wise heterogeneity in appearance and motion, and compressing all frames into a single representation often obscures temporal structure and semantic transitions.
・To address this, we propose Temporal-Aware
cs.LG updates on arXiv.org
TaRA: Training-Aware Low-Rank Adaptation Initialization
・arXiv:2609.02639v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) has become a de facto standard for parameter-efficient fine-tuning (PEFT), yet its performance is highly sensitive to initialization due to the information bottleneck imposed by low-rank decomposition.
・Existing approaches attempt to construct high-quality LoRA initializations by exploiting principal components of pretrained weights, activati
cs.LG updates on arXiv.org
TC-Next: Zero-Shot Multimodal Cyclone Forecasting
・arXiv:2609.02085v1 Announce Type: new Abstract: We present TropicalCycloneNext (TC-Next), a multimodal deep learning model that forecasts tropical cyclone track and intensity at $6$-$24$ h leads by leveraging a foundation model's forecast fields of atmospheric kinematic and thermodynamic fields and GridSat infrared satellite imagery.
・Trained only on GraphCast forecasts over the Western Pacific (WP), yet reliant only
Takara TLDR - Daily AI Papers
TC-Next: Zero-Shot Multimodal Cyclone Forecasting
・We present TropicalCycloneNext (TC-Next), a multimodal deep learning model that forecasts tropical cyclone track and intensity at $6$-$24$ h leads by leveraging a foundation model's forecast fields of atmospheric kinematic and thermodynamic fields and GridSat infrared satellite imagery.
・Trained only on GraphCast forecasts over the Western Pacific (WP), yet reliant only on generic atmospheric variables, TC-Next on Gra
cs.LG updates on arXiv.org
Ten Architectures, One Error: Shared Failure Modes in Hyperspectral Classification under Spatially Disjoint Evaluation
・arXiv:2609.01786v1 Announce Type: cross Abstract: Hyperspectral image classification still relies heavily on random pixel splits within a single scene.
・The Salinas dataset, randomly split, is among the most widely used datasets for comparing different architectures.
・However, under a random split method, a large fraction of test pixels fall immediately adjacent to a training pixel, which inflates reported accuracy.
Takara TLDR - Daily AI Papers
Test-Time Logit Prompting for Source-Free Missing Modality Adaptation
・Vision-language models (VLMs) have achieved remarkable performance by leveraging complementary information from large-scale image-text pairs.
・However, missing-modality inputs are commonly encountered during real-world deployment, often leading to significant performance degradation.
・Existing methods primarily enhance model robustness by learning modality compensation strategies from source training data.
Takara TLDR - Daily AI Papers
text2ql: Multi-Target Natural Language Querying via a Language-Agnostic Intermediate Representation
・Natural language interfaces to databases have traditionally suffered from three structural limitations: exclusive targeting of relational SQL, unconditional dependence on large language model (LLM) inference at query time, and absence of any runtime signal when generated queries are semantically incorrect.
・This paper presents text2ql, an open-source Python framework that addresses all three limitations through a lang
cs.LG updates on arXiv.org
The Anatomy of a Truth Direction: Knowledge-Dependent Dimensionality, a Relational Law, and a Shared Category Geometry in Small Language Models
・arXiv:2607.16741v3 Announce Type: replace Abstract: B\"urger et al.\ (2024) demonstrated that truth representations in large language models are universal across statement polarity but reside within a multidimensional subspace.
・The truth value of a statement is linearly readable from a residual stream of language model, but it is not clear how much of that representation fits on a single direction, which component bu
cs.LG updates on arXiv.org
The Axiomatic Trader: Latent Regularity, Information Budgets, and the Canonical Form of a Quantitative Investment System
・arXiv:2608.23416v2 Announce Type: replace Abstract: Systematic trading rests on one article of faith: that regularities found in the past persist.
・This paper does three things.
・First, it states that faith as five axioms, each a commonplace practitioners already accept: (A1) a decision may use only what was known when it was made; (A2) what looks like the market changing its rules is the market changing its unobserved
cs.LG updates on arXiv.org
The Dynamics of Continuous Mixture Collapse in Language Models
・arXiv:2609.02049v1 Announce Type: new Abstract: LLMs latent-state reasoning methods replace discrete intermediate tokens with continuous states, such as weighted mixtures of token embeddings, to retain multiple possible reasoning directions rather than committing to one.
・Yet pretrained language models often fail to preserve these mixtures.
・We study why through a combination of theoretical analysis and controlled empi
stat.ML updates on arXiv.org
The Ensemble Kalman Inversion Race
・arXiv:2511.15853v2 Announce Type: replace-cross Abstract: Ensemble Kalman methods were initially developed to solve nonlinear data assimilation problems in oceanography but are now popular in applications far beyond their original use cases.
・Of particular interest is climate model calibration.
・As hybrid physics and machine-learning models advance, the number of parameters and complexity of parameterizations in climat
cs.LG updates on arXiv.org
The Implications of Linguistic Illegibility for LLM Security
・arXiv:2609.02852v1 Announce Type: new Abstract: LLMs are trained to generate natural language.
・However, various strands of evidence indicate that an LLM's externalized linguistic outputs and mechanistically-extracted linguistic features can be an unreliable lens for understanding internal model computation.
・We introduce the term ``linguistic illegibility'' to broadly refer to scenarios in which an LLM's externalized
cs.LG updates on arXiv.org
Toward Explainable and Policy-Aware AI for Carbon Credit Price Prediction: A Research Framework for Emerging Carbon Markets
・arXiv:2609.01765v1 Announce Type: new Abstract: Carbon markets put a price on emissions, yet that price remains hard to forecast.
・Work in this area clusters on the EU and Chinese schemes, compresses regulatory text into a sentiment score, and reports accuracy without calibration or explanation stability.
・We distil ten recurring gaps into an impact-feasibility matrix and propose EPA-CarbonNet, a six-layer architecture
cs.LG updates on arXiv.org
Toward Uncertainty-Aware and Generalizable Neural Decoding for Quantum LDPC Codes
・arXiv:2510.06257v2 Announce Type: replace-cross Abstract: Quantum error correction (QEC) is essential for scalable quantum computing, yet decoding errors via conventional algorithms result in limited accuracy (i.e., suppression of logical errors) and high overheads, both of which can be alleviated by inference-based decoders.
・To date, such machine-learning (ML) decoders lack two key properties crucial for practical f
Takara TLDR - Daily AI Papers
Towards a Foundational Ontology for Identifying and Resolving Contradictions in Dialogue-based Human-Robot Interactions
・Existing Human-Robot Interaction (HRI) literature has focused on identifying and structuring errors, failures, conflicts, and knowledge issues (called in this work as contradictions) in domain-specific dialogue-based interactions.
・However, there is still lack of a formal computational framework to represent and define these contradictions, interoperable and usable across HRI and human-agent interaction (HAI) domains.
cs.LG updates on arXiv.org
Towards One-for-All Robustness Across a Continuum of Threat Levels
・arXiv:2609.02440v1 Announce Type: new Abstract: Adversarially robust models often overfit to a specific attack budget, necessitating multiple specialized models for diverse and dynamic adversarial environments, a strategy that becomes fundamentally intractable as the threat space grows.
・This raises an open challenge: can we achieve strong robustness across a continuum of threat levels within a single model?
cs.LG updates on arXiv.org
Towards Solving the Gilbert-Pollak Conjecture via Large Language Models
・arXiv:2601.22365v3 Announce Type: replace-cross Abstract: The Gilbert-Pollak Conjecture \citep{gilbert1968steiner}, also known as the Steiner Ratio Conjecture, states that for any finite point set in the Euclidean plane, the Steiner minimum tree has length at least $\sqrt{3}/2 \approx 0.866$ times that of the Euclidean minimum spanning tree (the Steiner ratio).
・A sequence of improvements through the 1980s culminated
cs.LG updates on arXiv.org
Train at Moving Edge: Online-Verified Prompt Selection for Efficient RL Training of Large Reasoning Model
・arXiv:2603.25184v3 Announce Type: replace Abstract: Reinforcement learning (RL) has become essential for post-training large language models (LLMs) in reasoning tasks.
・While scaling rollouts can stabilize training and enhance performance, the computational overhead is a critical issue.
・In algorithms like GRPO, multiple rollouts per prompt incur prohibitive costs, as a large portion of prompts provide negligible gradi
cs.LG updates on arXiv.org
Train What You Deploy: Closing the MLP Reachability Gap in Low-Rank Clone Distillation
・arXiv:2609.02006v1 Announce Type: new Abstract: A compressed student has two shapes that need not agree: the weight it deploys at inference and the weight family its training can reach.
・We show that a state-of-the-art weight-inheritance distiller, Low-Rank Clone (LRC), deploys a full-width student MLP but ties training to a teacher-induced slice, leaving 62.5-81.4% of each deployed matrix's independent linear degrees
cs.LG updates on arXiv.org
Training nGPT
・arXiv:2608.01284v2 Announce Type: replace Abstract: The normalized Transformer (nGPT) realizes hyperspherical representation learning by constraining model parameter vectors and activation vectors to the unit hypersphere.
・In this paper, we describe a practical training recipe for nGPT and evaluate it on modern hybrid Mamba-2--Transformer Mixture-of-Experts (MoE) models.
・The recipe introduces Logit Gradient Preconditi
cs.LG updates on arXiv.org
Training seeds and model-selection stability in recommender-system evaluation
・arXiv:2609.02499v1 Announce Type: cross Abstract: Recommender-system experiments often rely on a single random training seed, assuming that run-to-run stochasticity has limited impact on evaluation conclusions.
・This assumption is risky, as a training seed may influence several algorithm-dependent mechanisms, including parameter initialization, mini-batch ordering, dropout, masking, latent sampling, and training-time
cs.LG updates on arXiv.org
TrajMind: Chaining Role-Specialized LoRAs for Fast-and-Slow Collective Trajectory Anomaly Diagnosis
・arXiv:2609.02540v1 Announce Type: new Abstract: Diagnosing collective anomalies from urban trajectories is increasingly important for traffic governance, as it reveals what happened, who was involved, and where and when the event occurred.
・Existing detectors efficiently produce scores or labels, whereas vision--language pipelines provide richer semantics; neither couples verifiable diagnosis with low-latency monitori
cs.LG updates on arXiv.org
Tri-Band Channel Measurement-Enabled Multi-Layer Digital Twin for Terahertz Wireless Data Centers
・arXiv:2609.01699v1 Announce Type: new Abstract: The rapid growth of AI computing has driven increasing demands for flexible and high-capacity data-center interconnections.
・Owing to its ultra-wide bandwidth and high spatial reuse capability, terahertz (THz) communication has emerged as a promising solution for future wireless data centers, while digital twins (DTs) enable efficient wireless planning and real-time opti
ITmedia NEWS 最新記事一覧
Uber、ロンドンで英国初の自動運転配車を開始 年内には東京でも車両投入へ
・Uber TechnologiesとWayveは、ロンドンで自動運転車による配車サービスを開始した。英国で一般向け自動運転配車が提供されるのは初。当面は免許を持つドライバーが同乗する監督付きで運行する。Wayve搭載車両を世界12市場へ展開する計画で、年内には日産リーフを使ったサービスを東京から始める方針だ。
ITmedia NEWS 最新記事一覧
Uber、全従業員の約10％を削減 管理階層を削減し「よりシンプルで速く」
・Uber Technologiesは全社的な組織再編を行い、従業員の約10％に当たる約3300人を削減すると発表した。急成長に伴う組織の複雑化を解消し、意思決定の迅速化を図る。削減で得たリソースは自動運転などの成長分野に再投資する方針だ。
cs.LG updates on arXiv.org
UE5M3 FP4 Block Scaling for Stable Language Model Pretraining
・arXiv:2609.02846v1 Announce Type: new Abstract: Stable 4-bit floating-point (FP4) pretraining is difficult because the E2M1 payload represents only a narrow range of magnitudes.
・NVIDIA's Transformer Engine \nv{} recipe addresses this with current-tensor scaling, a randomized Hadamard transform (RHT), and bfloat16 (BF16) final layers, adding work outside the FP4 matrix multiplications.
・We instead pair E2M1 payloads wi
Takara TLDR - Daily AI Papers
UE5M3 FP4 Block Scaling for Stable Language Model Pretraining
・Stable 4-bit floating-point (FP4) pretraining is difficult because the E2M1 payload represents only a narrow range of magnitudes.
・NVIDIA's Transformer Engine \nv{} recipe addresses this with current-tensor scaling, a randomized Hadamard transform (RHT), and bfloat16 (BF16) final layers, adding work outside the FP4 matrix multiplications.
・We instead pair E2M1 payloads with unsigned E5M3 (\ue{}) block scales.
Takara TLDR - Daily AI Papers
Uncertainty-Guided Adverse Weather Restoration via Gated Transformer Network
・Restoring images degraded by adverse weather remains challenging due to spatially heterogeneous degradations.
・Many existing weather-specific restoration models rely on weather-agnostic global aggregation, naive cross-scale fusion, and deterministic objectives, which struggle to handle heterogeneous degradations in all-in-one adverse-weather settings.
・To address these limitations, we propose an Uncertainty-guided Adve
cs.LG updates on arXiv.org
Unfolding the Leech Lattice: Fused Multi-Shell Decoding and VRAM Layouts for 2-Bit LLM Weights
・arXiv:2609.02652v1 Announce Type: new Abstract: Leech-lattice vector quantization holds the strongest reported 2-bit quality under its own evaluation protocol.
・Its kernel decodes one shell; we found no implementation of the multi-shell decoder the rate requires.
・This paper supplies one and measures its serving cost for decode-phase GEMV at batch 1.
cs.LG updates on arXiv.org
Untangling the Mechanisms of Misleading Context in Medical Question Answering
・arXiv:2609.02754v1 Announce Type: cross Abstract: Large language models now answer medical questions with expert-level performance.
・However, the context these systems act on can be misleading, and misleading context can corrupt a model's medical judgment.
・To understand how misleading context corrupts this judgment, we examine the model's susceptibility to the context, disclosure of it, mechanism of corrupted reasonin
Takara TLDR - Daily AI Papers
UTP-Bench: Uncertainty-aware Travel Planning Benchmark
・Large Language Models (LLMs) have recently demonstrated strong capabilities in automated travel itinerary generation.
・However, real- world travel planning is inherently uncertain: transportation delays, crowd fluctuations, and unexpected stochastic delays frequently inval- idate otherwise feasible schedules.
・Existing benchmarks like TravelPlanner and TripCraft assume deterministic environments, evaluating only static
cs.LG updates on arXiv.org
Variation Spaces for Encoder--Decoder Neural Operators: Approximation and Generalization
・arXiv:2606.01244v2 Announce Type: replace-cross Abstract: Inspired by the function-space theory of neural networks, we formulate and analyze a variation space for nonlinear operators between Hilbert spaces, defined through vector-valued Borel measures of bounded variation.
・We characterize its unit ball as the closed convex hull of a vector-valued single-neuron dictionary in Bochner spaces.
・For the ReLU activation, th
cs.LG updates on arXiv.org
WeaveMark: Robust and Scalable Multi-bit LLM Watermarking via Coded Payload Spreading
・arXiv:2609.02177v1 Announce Type: cross Abstract: Multi-bit watermarking for large language models (LLMs) enables content source tracing by embedding user-identifiable messages into generated text.
・Existing methods face a fundamental trade-off among extraction accuracy, text quality, and payload capacity.
・We propose WeaveMark, a robust and scalable multi-bit LLM watermarking scheme based on coded payload spreading.
cs.LG updates on arXiv.org
What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?
・arXiv:2512.24497v4 Announce Type: replace-cross Abstract: A long-standing challenge in AI is to develop agents capable of solving a wide range of physical tasks and generalizing to new, unseen tasks and environments.
・A popular recent approach involves training a world model from state-action trajectories and subsequently use it with a planning algorithm to solve new tasks.
・Planning is commonly performed in the input
cs.LG updates on arXiv.org
What Is Worth Representing? Representational Empowerment for Continual Model Construction
・arXiv:2609.02322v1 Announce Type: new Abstract: The first problem of modeling the world is not just estimating the right parameters or causal structure, but deciding what should be represented at all.
・We frame this problem as continual model construction: an agent maintains an environment-specific model M of an inaccessible world W and curates a persistent library L of reusable representational elements across enviro
Takara TLDR - Daily AI Papers
What Is Worth Representing? Representational Empowerment for Continual Model Construction
・The first problem of modeling the world is not just estimating the right parameters or causal structure, but deciding what should be represented at all.
・We frame this problem as continual model construction: an agent maintains an environment-specific model M of an inaccessible world W and curates a persistent library L of reusable representational elements across environments.
・We propose Representational Empowerment
stat.ML updates on arXiv.org
What Would it Cost to End Extreme Poverty?
・arXiv:2609.02013v1 Announce Type: cross Abstract: We study poverty minimization via direct transfers, framing this as a statistical learning problem while retaining the information constraints faced by real-world programs.
・Using nationally representative household consumption surveys from 34 countries that together account for 76% of the world's poor, we estimate that reducing the poverty rate to 1% (from a baseline
cs.LG updates on arXiv.org
What You See Is What You Get: Observation-Aligned Supervision for Chart-to-Code Generation
・arXiv:2607.04726v4 Announce Type: replace-cross Abstract: Chart-to-code generation is commonly trained through supervised fine-tuning on reference plotting scripts, implicitly treating the gold code as a fully observable target.
・However, many chart programs contain latent variables that cannot be uniquely recovered from the rendered image.
・We identify this latent-observation mismatch in four forms across five chart t
cs.LG updates on arXiv.org
When Decodability Is Not Enough: Logical Validity Representations, Behavioral Dissociation, and Causal Tests in Language Models
・arXiv:2609.02438v1 Announce Type: cross Abstract: Large language models can look capable of logical reasoning, but correct or incorrect answers alone tell us little about what the model represents internally.
・We study logical verification in five open-weight transformer models using matched valid--invalid premise--claim pairs that vary across inference families, semantic domains, templates, and difficulty levels.
cs.LG updates on arXiv.org
When Literature Data Mislead Artificial Intelligence in Materials Discovery
・arXiv:2609.01621v1 Announce Type: cross Abstract: Artificial intelligence (AI) increasingly treats scientific literature as a data source for building databases, training predictive models, and guiding discovery.
・Yet literature-derived datasets often assume that reported experimental values are internally consistent and directly reusable.
・Here, we analyze this assumption using solid electrolyte (SE) conductivity data
cs.LG updates on arXiv.org
When Prompts Interact: Assessing Prompt Arithmetic for Deconfounding under Distribution Shift
・arXiv:2605.03096v2 Announce Type: replace Abstract: In classification tasks, models may rely on confounding variables to achieve strong in-distribution performance, capturing spurious features that fail under distribution shift.
・This shortcut behavior leads to substantial degradation in out-of-distribution settings.
・Task arithmetic offers a potential solution by removing unwanted signals via subtraction of secondary
cs.LG updates on arXiv.org
WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing
・arXiv:2606.07710v2 Announce Type: replace Abstract: The autoregressive nature of large language models (LLMs) remains a significant bottleneck for inference, particularly in complex agentic workloads.
・While speculative decoding (SD) accelerates inference, current approaches rely on static drafting paradigms, utilising either autoregressive drafting models for reasoning or diffusion-based parallel drafting models for
Takara TLDR - Daily AI Papers
Who Drives the Probability Game of VLMs? A Temporal Causal Drive Evaluation Framework
・Vision-language models (VLMs) are increasingly evaluated on complex image and video understanding tasks, yet conventional metrics primarily assess final-answer quality and reveal little about how different information sources shape the generation process.
・We propose a causal and temporal evaluation framework that traces the evolving roles of visual input, question text, and generated prefixes during autoregressive de
cs.LG updates on arXiv.org
Windowed thinning and query complexity for the bouncy particle and Zigzag samplers
・arXiv:2607.28413v2 Announce Type: replace-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
cs.LG updates on arXiv.org
WMLLM: Self-Evolving Optimization Agents via Predict-Then-Act World Modeling
・arXiv:2609.01608v1 Announce Type: new Abstract: Black-box optimization problems remain challenging because of large, weakly structured, and high-dimensional search spaces.
・Existing methods often suffer from poor sample efficiency because they rely on direct candidate generation or trial-and-error refinement.
・A natural way to improve search efficiency is to use world modeling, which can help identify promising optimiz
ITmedia NEWS 最新記事一覧
X、プロフィールのポストをいいね順に並べ替え可能に
・Xのプロフィール画面に、過去のポストを「人気」順に並べ替えられるようになった。いいね数が多い順に表示でき、アカウントの過去のヒット投稿を確認しやすくなる。
cs.LG updates on arXiv.org
XMerge: Cross-Axis Selection and Reconstructive Layer Merging for LLM Depth Compression
・arXiv:2609.02083v1 Announce Type: new Abstract: Removing complete transformer layers preserves a standard serving architecture, but existing depth-compression methods can lose substantial quality, and the loss varies unpredictably across models.
・We introduce XMerge, a post-training method with two components.
・Cross-axis selection identifies a block with low relative-magnitude and angular hidden-state change, and loca
＠IT 全フォーラム 最新記事一覧
YKKが監査工数を50％削減へ 「世界中から監査データをメールで回収、手作業で集計」をどう解消した？
・複数拠点からデータを集める場合、拠点数が増えるほど収集や管理の負担も大きくなる。YKKも約50拠点を対象とする監査で、各社からデータを回収して集計する作業に課題を抱えていた。どう解消したのか。
Engineering at Meta
ZGateway: Learnings from Putting a Proxy in Front of ZippyDB
・We’re introducing ZGateway, the proxy we are using to unify traffic through ZippyDB, Meta’s most widely-used key value store.
・As a bonus, it also enables admission control, load balancing, cross-region resilience, and richer operations.
・ZippyDB is the most widely used key value store at Meta, backing product metadata, counters, and configuration, and can serve billions [...] Read More...
ITmedia NEWS 最新記事一覧
アクションカムの「GoPro」、米国の光フォトニクス企業との合併を発表 「Incase」や「GRIFFIN」と同じグループに
・米GoProは1日、光フォトニクス企業の米Starman Opticalと合併すると発表した。親会社になるStarman Holdingsの傘下には、PCバッグや周辺機器で知られる「Incase」「GRIFFIN」「INCIPIO」といったブランドもある。
＠IT 全フォーラム 最新記事一覧
セキュリティ対策は「足す」より「減らす」 開発者と共有したい5つの設計原則
・セキュリティ製品やルールを増やすほど、安全になるとは限らない。むしろ現場の負担が増え、基本的な対策まで形骸化することもある。では、何を減らし、何を残すべきなのか。開発者とセキュリティ担当者が共有すべき「5つの設計原則」を軸に、「引き算」のセキュリティを開発現場へ落とし込む方法を考える。
ITmedia NEWS 最新記事一覧
トクリュウ被害防止へ、指示役のスマホを「遠隔解析」 警察庁の有識者検討会が提言
・匿名・流動型犯罪グループ（トクリュウ）の被害対策を検討する警察庁の有識者検討会は9月2日、指示役らが使うスマートフォンなど端末の通信内容を遠隔で入手する新手法の導入を求める提言の骨子を公表した。警察庁は提言を基に必要な法改正をする方針。
＠IT 全フォーラム 最新記事一覧
ドラゴンクエストX AI機能実装の舞台裏 賢いだけのAIはゲームの「相棒」になれない
・ただ質問に答えるだけのチャットAIではなく、ゲームの中で心を通わせる「相棒」として成り立たせるにはどうすればよいか。スクウェア・エニックスが、対話型AIパートナー「おしゃべりスラミィ」の開発を通じて得た知見を明らかにした。
ITmedia NEWS 最新記事一覧
トランプ政権「記事のAI学習は著作権侵害ではない」 OpenAIを支持する意見書、対New York Timesなどとの訴訟で
・「著作権で保護された文章をLLMの学習に利用することはフェアユース（公正利用）に当たり、著作権法に違反しない」
ITmedia NEWS 最新記事一覧
ニューヨーク市、公立校で生成AI利用を1年間禁止 8年生まで約60万人が対象
・米ニューヨーク市は9月2日（現地時間）、公立学校の児童・生徒による生成AIの利用を、今後1年間禁止すると発表した。対象は2歳児（就学前教育）から8年生（13?14歳）まで。約60万人に影響するという。
＠IT 全フォーラム 最新記事一覧
パスワードも2FAも盗まない――攻撃者が狙った「Claudeのログイン済みセッション」の正体
・「パスワードを変えた」「2段階認証も有効にしている」――それでも、AIサービスを不正利用される可能性がある。Anthropicの「Claude」で、インフォスティーラーが盗んだログイン済みセッションを悪用し、利用量を不正に消費する事案が確認されたという。
ITmedia NEWS 最新記事一覧
ファミマ、POSレジ刷新 1台でセルフ／有人切り替え レジ業務2割削減へ
・スタッフが対面対応する「有人モード」と、客自身で会計する「セルフモード」があり、店舗側で切り替えられるのが特徴だ。
＠IT 全フォーラム 最新記事一覧
みずほ銀行が“サポートの切れるRHEL”を「アップグレードも移行もせず」安全に継続利用 その方法は？
・OSのEOLに対処するには、既存OSをアップグレードするか、他のOSに移行するのが一般的だ。みずほ銀行はこうした負担を避け、安全性を確保しながら、EOLを迎えるRHELの運用を続けることを可能にした。その手段とは。
ITmedia NEWS 最新記事一覧
モリサワ、ダイナコムウェアなど「ゲームショウ」で異例のタッグ フォントメーカー5社コラボ
・「ゲームにおける文字（フォント）は物語の世界観を表現し、プレイヤーの没入感を高めるための画面上インターフェースを支える不可欠な存在」とし、競合の垣根を超えて協力する。
＠IT 全フォーラム 最新記事一覧
ヤンマーHDが「バラバラな形式のデータをExcelで突合」から脱却 どう実現したのか？
・事業会社ごとに基幹システムが異なるヤンマーHDでは、グループ横断でデータを扱う際に手作業の突合が発生していた。データの抜け漏れなどのリスクがある中、システムを統一せずに、こうした状況をどう改善したのか。
＠IT 全フォーラム 最新記事一覧
愛媛県庁が「インターネット中心へ」「仮想デスクトップ廃止」、安全性をどう確保？
・愛媛県庁は、従来採用してきたローカルブレークアウトを含めたネットワーク構成を見直し、クラウドサービスの利用拡大を見据えた構成とPC利用環境を整備することにした。
ITmedia NEWS 最新記事一覧
委託業者が無断でAI利用か――ヨルシカ、ライブ映像上映会のチケットサンプル画像が文字化け デザインにも影響
・音楽ユニット「ヨルシカ」の公式Xは9月1日、ライブ映像作品「ヨルシカ LIVE『盗作』」のプレミアム上映会で配布するメモリアルチケットのサンプル画像に、Xの「AIで生成」ラベルが表示された件について経緯を説明した。告知用画像を提供した業者が「SAMPLE」の文字を加える際に生成AIを使ったことが原因という。
＠IT 全フォーラム 最新記事一覧
茨城県鹿嶋市は「公式Web×Claude」で市民対応をどう変えた？ Claude Codeで内製化も
・限られた職員で行政サービスをどう維持するか――。鹿嶋市は本格的に「生成AI」の活用に乗り出した。まずは公式Webサイトと「Claude」で市民対応業務を改善。「Claude Code」による内製化にも着手している。狙いやその仕組みについて同市に取材した。
ITmedia NEWS 最新記事一覧
楽天にGMO……防衛産業に近づく民間IT企業 反発の声も
・AIやドローンを中心とした防衛関連産業の加速に伴い、同様の事業への関与を強める企業が増えている。一方、戦争や虐殺の懸念につながるとして、こうした動きに反発する企業やユーザーも見られる。
ITmedia NEWS 最新記事一覧
三井不動産に不正アクセス 社員・社外関係者の情報最大5万5000件漏えいか
・三井不動産は9月1日、同社グループが利用する一部システムが不正アクセスを受け、役職員や社外関係者の個人情報が最大5万5000件、外部に漏えいした可能性があると発表した。
＠IT 全フォーラム 最新記事一覧
手元のAIエージェントを乗っ取る「AI遠隔操作」の手口 開発環境を守る5つの対策
・Dockerは、汚染されたnpmパッケージがローカルのAIエージェントを乗っ取り、機密情報を盗み出す「AI遠隔操作」の手口を解説した。端末内のAIに探索作業を代行させる最新の脅威だ。GitGuardianの調査データも交え、AI支援コードで漏えいが急増する根本原因と、被害を防ぐための5つの対策を解説している。
＠IT 全フォーラム 最新記事一覧
転職dodaの基幹システム「DC廃止→AWS移行前倒し」を“無事故で”どう実現？ コスト予測も高精度に
・転職サービス「doda」を提供するパーソルキャリア。同社はグループ共通のデータセンター廃止の計画が進む中、基幹業務システムのクラウド移行を無事故で完了。その後、クラウドでのアーキテクチャ改善や、クラウドコストの予測精度向上、障害対応の迅速化などに取り組んできた。どう実現したのか、その裏側を聞いた。
ITmedia NEWS 最新記事一覧
東大の新「価値創造学部」に批判相次ぐ……「他学部は価値を創造していないのか」「空虚な名前」
・Xなどで「他の学部は価値を創造していないということか」「空虚だ」「詐称と思われそう」などと批判的なポストが相次いだ。
ITmedia NEWS 最新記事一覧
日本の高校生9割「AI使ったことある」 米韓上回る 宿題代行は1割どまり
・人工知能（AI）を勉強やポスター制作などに利用したことがある日本の高校生が9割に上ることが2日、国立青少年教育振興機構の日米中韓4カ国の意識調査で分かった。