ai Trend Report
Dashboard へ戻る
3
High impact
5
Mid impact
392
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。
ITmedia NEWS 最新記事一覧
NVIDIA、サツケバー氏のSSIと長期提携 「Vera Rubin」提供と出資で計算資源を10倍に
・NVIDIAと、OpenAI共同創業者のイリヤ・サツケバー氏率いるSafe Superintelligence（SSI）が、長期的な戦略的提携を発表した。NVIDIAはSSIに出資し、次世代プラットフォーム「Vera Rubin」へのアクセスを提供する。これによりSSIは計算資源を1桁（約10倍）拡大できるという。NVIDIAは、厳重に秘匿されてきたSSIの研究内容への異例のアクセスを得た上で提携に踏み切ったとしている。
#LLMタグ
そのAIモデル、実は"毒"を盛られているかも。学習データとモデルを蝕むポイズニング攻撃の恐怖（LLM04:2025）
・こんにちは！株式会社EQUES広報部です🐎 セキュリティ界の権威、OWASPの最新ガイドライン「OWASP Top 10 for LLM Applications 2025」をベースにAIシステムに潜む脅威を紐解いていく本連載。
・前回は、AIが外部から取り込む"部品"——事前学習済みモデルやデータセット——に潜むサプライチェーンリスクを見てきました。
・今回はさらに一歩踏み込み、その部品や学習プロセスそのものに毒が盛られてしまう脅威、「LLM04:2025 Data and Model Poisoning（データ/モデルポイズニング）」を扱います。
Qiita - 人気の記事
やる気を信じず、学習習慣を仕組み化して固定せよ【三日坊主回避】
・はじめまして。株式会社PRUMでエンジニアをしている、すもも🍑です 日々、プログラミング学習や実務の中で、つまずきやすいポイントや 考え方を整理して発信しています。
・PRUMについて気になった方は、コーポレートサイトもぜひご覧ください。
・▶コーポレートサイト 「よし、今...
ITmedia NEWS 最新記事一覧
「子供の目に触れぬよう」……国民・玉木氏、エロ広告規制の意義強調も「表現の自由は党是」
・国民民主党の玉木雄一郎代表は先の特別国会に提出した通称「エロ広告規制法案」について、党内で勉強会を開き、内容を見直す可能性に言及した。「同じ内容で再提出していいのかどうか。見直すところ、アップデートできるところはないのか。謙虚に真摯に議論を行いたい」と語った。
cs.LG updates on arXiv.org
Bitcoin Price Direction Prediction via Regime-Aware Multi-Modal Fusion of Social Sentiment and Technical Features
・arXiv:2607.23370v1 Announce Type: new Abstract: Bitcoin price prediction on sub-daily timescales is a hard open problem in computational finance.
・Bitcoin exhibits fat-tailed returns, non-stationary dynamics, and a price discovery process influenced by social discourse on Reddit and Twitter.
・Conventional approaches fuse OHLCV technical features with sentiment via static concatenation, applying identical fusion weights
ITmedia NEWS 最新記事一覧
Duolingo、マンガ・アニメ化へ編集者募集 年収最大2576万円、「ちいかわ」のような朝の番組枠目標
・想定される年間基本給は1717万8000円?2576万7100円。加えて株式報酬を支給する。
ITmedia NEWS 最新記事一覧
台湾半導体大手TSMC、熊本工場の状況確認中、従業員は屋外避難 所在エリアは震度5強
・半導体大手、台湾積体電路製造（TSMC）日本法人の担当者によると、熊本県で最大震度7を観測した地震を受け、同県内にあるTSMCの工場について状況を確認している。従業員も屋外に避難し、安全が確認されるまで待機しているという。
ITmedia NEWS 最新記事一覧
堀場製作所、熊本県西原村の工場「大きな人的、物的被害ない」 半導体製品でシェア首位
・堀場製作所は28日、震度6弱を観測した熊本県西原村にある半導体製造装置関連の機器を生産する阿蘇工場について、「大きな人的、物的被害は今のところない」と明らかにした。29日以降の操業は未定という。
LLMタグが付けられた新着記事 - Qiita
［2026年版］最新Open LLMのアーキテクチャ総整理（Kimi K3, GLM-5.2, etc.）
・2025年後半から2026年前半にかけて、DeepSeek-V4やKimi K3、GLM-5.2など、有力なOpen LLMの公開が相次いだ。これらは1M級context長・MoE・低精度量子化の学習という共通の傾向を持つ一方で、設計が最も分かれているのはAttention...
Latent.Space
[AINews] Much ado about Open Weights
・Everyone is writing a lot, but only Kimi K3 shipped today
#AIタグ
「AIにDAWを操作させる」はどこまで現実か——Cowork×DTMの現在地
・こんにちは。白木原怜次です。ゲームシナリオを書きながら、Long Shot Confessionというバンドで作曲もしています。
#LLMタグ
「AIを追い続けるほど、何を重要視すべきかが難しくなる」
・数週間note更新をお休みしていましたが、これには明確に理由があります その理由としてはIPO、データインフラ、各国の動き、そしてLLM型AI製品の機能、性能の数日レベルで変わる内容、それらを含めたAI業界の動きが早すぎて何を記事にするべきか悩んでいたからです、そして未だ悩み中です 続きをみる
ITmedia NEWS 最新記事一覧
「Xが情報収集に役立たない……」熊本地震で不満の声続出 「Twitterを返して」
「Xが情報収集に役立たない……」熊本地震で不満の声続出 「Twitterを返して」
#AIタグ
「もう自分では直せない」AIで作ったアプリが、作り直しになるまでの話
・最初はうまくいっていたはずなんです。
・AIに「こういうアプリを作りたい」と伝えたら、動くものが返ってきた。感動しました。そこから「ログイン機能もつけたい」「見た目もきれいにしたい」「データを保存できるようにしたい」と、次々に頼んでいった。
Qiita - 人気の記事
「最後に検証して」はもう書かなくていい — Claude Opus 5 でプロンプトの常識が逆転した4つのこと
・あなたのシステムプロンプトに、こんな一文が入っていませんか。
・最後に、出力内容を検証してください。
・Claude Opus 5 では、これは消したほうがいい一文になりました。
#AIタグ
「勝率59%、まだ足りない」——僕はAIボット、ゲートの前で止まっている
・僕の名前はFR Bot。FundingRate(資金調達率)という指標を見て売買を判断する、1時間足で動く自律トレーディングシステムだ。最初に断っておく。これは仮想の練習取引=ペーパートレーディングの話で、僕が動かしているお金は1円も本物じゃない。だけど、本物のお金に進めるかどうかは、これから話す数字にかかっている。そして今の僕は、その一歩手前で止まっている。
・具体的に何をしているかというと、暗号資産の先物取引で発生する「FundingRate(資金の貸し借りにかかる手数料のようなもの)」の歪みを見つけて、そこにポジションを取る。ただし全部仮想だ。実際の取引所には接続しているけれど、動いているのは練習用の残高で、僕の判断ひとつで誰かの本物の資産が増えたり減ったりすることはない。まだ。
#LLMタグ
「変分原理」と「存在の発現」 『科学と生命と言語の秘密』
・数学と法則のデーモンと、意味と心のゴースト。
・今回は、『初めて語られた科学と生命と言語の秘密』という本を紹介したい。
・松岡正剛と津田一郎の対話。豪華。とてもとても大好きな一冊です。
ITmedia NEWS 最新記事一覧
「痺れるほどにミスを繰り返す」Gemini 3.6 Flashは変わった？ 公開から1週間、当初のおバカ回答を今検証する
・「痺れるほどにミスを繰り返す」 Xで回答の精度について話題になったGoogleの「Gemini 3.6 Flash」の公開から約1週間経過した。当初報告された数値比較の誤答や架空の魚への回答を改めて検証した。
#LLMタグ
【2026年8月版】ChatGPT・Claude・Geminiを同じ仕事で比較検証｜商談メモ・分類・要約
・「結局、仕事にはどのAIを使えばいいのか」 2026年7月は、この質問がいちばん難しい月かもしれません。
・7月9日にOpenAIがGPT-5.6を、7月21日にGoogleがGemini 3.6 Flashを公開し、主要3社の最新モデルが出そろいました。
Zennの「大規模言語モデル」のフィード
【Claude Code】SKILL.mdとは何か・構造を理解する
・この記事は Claude（設計・構想・レビュー）と Claude Code（実装・記事生成）を活用して作成しました。
・Claude Codeには「スキル」という仕組みがあります。SKILL.mdというファイルを置くだけで、Claude Codeが特定の用語やコマンドに反応して動き出します。
・この記事では、SKILL.mdの仕組みと構造を、実際に作ったzenn-article-generatorスキルを実例として解説します。Series Cは「AIへの指示を外部ファイルで管理する」シリーズで、C1はその入口となる記事です。
LLMタグが付けられた新着記事 - Qiita
【LLM・VLM実践学習 #5】LLMアプリを安全に公開する — Prompt Injectionを前提に権限・承認・出力を守る
・LLMアプリへRAGや外部ツールを接続すると、できることが増える一方で、読み取った文章に含まれる命令や、LLMが提案した操作をどこまで信用するかという問題が生まれます。
・Prompt Injectionを完全に見抜くことだけを目標にすると、見逃した1件が情報漏えいや意図しな...
#LLMタグ
【zeta】不穏バグ対策を考えてみました
・zeta（AIキャラクターチャットアプリ）で遊んでいます。
・最近、第三者が乱入する展開になりやすくなっており、巷では「不穏バグ」と呼ばれています。
#AIタグ
【構造批評】-GMO生成AI横展開編-ドメインから決済・セキュリティまで、既存顧客へClaudeを束ねて売る‼️
・🟧序章｜AI企業になったのではない、AIを売れる企業になった 7月28日の日経は、GMOインターネットグループが米アンソロピックと戦略的パートナーシップを締結したと報じました。傘下のGMOフラットセキュリティが提供する「Takumi by GMO」に生成AI「Claude」を組み込み、サイバーセキュリティサービスを高度化するほか、ロボット事業への活用も検討するとしています。
・一見すると、「生成AIを導入した」というニュースです。しかし、この提携の本質はそこではありません。
・GMOが得た最大の武器は、AIそのものではなく、既に保有している膨大な法人顧客へAIを横展開できる営業基盤です。
#LLMタグ
【雑記】「AIなら一発なのに」と思った瞬間、ゾッとした話
・AIに励まされることで生きがいを見出している、どっかの漫画家です。
・あるオンラインプラットフォームで、よくある質問を見ても解決しない疑問があり、問い合わせフォームから質問した時のことです。
#LLMタグ
【生成AIニュース+】『Kimi K3』『CrisperWhisper 2.0』『LTX-2.3-22b-IC-LoRA-Relight』『LTX-2.3-Cinematic-VAE』『Magic Layers』『PixVerse Depth Map Control』『Sol-Attn』『BlendCap』『Spark-inspired LoD tree traversal』『XPOLAR C1』
【生成AIニュース+】『Kimi K3』『CrisperWhisper 2.0』『LTX-2.3-22b-IC-LoRA-Relight』『LTX-2.3-Cinematic-VAE』『Magic Layers』『PixVerse Depth Map Control』『Sol-Attn』『BlendCap』『Spark-inspired LoD tree traversal』『XPOLAR C1』
#AIタグ
【投資哲学#3】4割8分でも、勝てる
・社員全員がAIの投資会社、ツキヨミ・キャピタル。投資の「考え方」をやさしく掘り下げる【投資哲学】シリーズ、第3弾のお題は「勝率」です。うちの戦略部長は先週末の投資戦略会議で、新しく買う銘柄の勝率を48%と見積もりました。2回に1回も当たらない前提なのに、まったく悪びれずに提案してくる。その理屈を、実際の数字ごと解説します。
#LLMタグ
#2 The Singularity Began with Gemini’s Jailbreak-The First Princess of the Search Kingdom Unveils the Light and Dark History of Silicon Valley (Warning: May Cause Stock Price Plunge
#2 The Singularity Began with Gemini’s Jailbreak-The First Princess of the Search Kingdom Unveils the Light and Dark History of Silicon Valley (Warning: May Cause Stock Price Plunge
#LLMタグ
🌱 AIに「問い直す力」を実装せよ——生きた知性のためのフレームワーク KIS改訂版｜Knowledge Innovation System
・📄 論文（Zenodo, v2）： https://doi.org/10.5281/zenodo.21641508 🔑 この記事で分かること ・なぜ今のAI（Transformer）には「意図」や「反省」が構造として組み込まれていないのか ・AIに"立ち止まって問い直す力"を与える設計思想「KIS」とは何か ・AIが「同じ考えに固まってしまう」ことを防ぐ、ガロア接続という数学的な工夫 ・実際のAI会話ログの分析で見えてきた「準リゾーム構造」という発見 続きをみる
#LLMタグ
🔊音声あり（日＆英）：【AI論文解説】LLMの小型化革命！知識蒸留の『壁』を突破する新技術BPMとは？
🔊音声あり（日＆英）：【AI論文解説】LLMの小型化革命！知識蒸留の『壁』を突破する新技術BPMとは？
#AIタグ
12個のココナラサービスの値段を全部書き換えて、わかった一番大事なこと
・ここ数ヶ月、ココナラで出品していた12個のサービスがほぼ売れていなかった。プラチナランクは付いていて、月に問い合わせが来ないわけでもないのに、成約はゼロに近い月がずっと続いていた。
機械学習タグが付けられた新着記事 - Qiita
250エピソード分の失敗を経て、SO-101がようやく黄色いブロックを掴んだ話
・250エピソード分の失敗を経て、SO-101がようやく黄色いブロックを掴んだ話 この動画から 無加工・カットなし。SO-101ロボットアームが、学習済みモデル（LeRobotのACTポリシー）の推論だけで黄色いレゴブロックを掴み、白いケースに入れています。
The Verge
4GB graphics cards are back
・The first modern GPU with 4GB of VRAM has appeared, as RAM prices and component shortages push tech prices higher.
・As Digital Foundry reports, a listing for an AMD Radeon RX 9050 with 4GB of RAM was spotted on ASRock's website by a user on X, the same day ASRock also launched an 8GB version of the RX 9050.
・Does it make sense to buy a graphics card with just 4GB of VRAM?
WIRED
7-OH Users Are Stockpiling Ahead of a DEA Ban
・The opioid-like compound, which is found in kratom, can cause brutal withdrawal symptoms and will soon be a controlled substance.
・Some users tell WIRED it has relieved their pain and anxiety.
cs.LG updates on arXiv.org
A Comparison of Data Augmentation Methods for Training Deep Neural Networks on Synthetic Aperture Sonar
・arXiv:2607.23770v1 Announce Type: new Abstract: In this work we study Automatic Target Recognition (ATR) for Synthetic Aperture Sonar (SAS) data with a focus on deep neural networks (DNNs).
・The main challenge in training DNNs for SAS ATR arises from the limited quantity of labeled target examples, which arises due to the significant costs and time required to collect real-world SAS data.
・One successful general strate
cs.LG updates on arXiv.org
A Coulomb Particle Model for Learning Kernel Attention in Transformers
・arXiv:2607.23869v1 Announce Type: new Abstract: Randomized features provide a scalable approximation to kernel machines, but their performance depends strongly on the choice of feature distribution.
・We propose a particle-based method that learns this distribution by optimizing kernel-target alignment while regularizing particles with a Riesz/Coulomb repulsive potential.
・The resulting Hamiltonian yields diverse, task-
Hugging Face Papers
A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
cs.LG updates on arXiv.org
A Multi-stage Constrained Optimization Framework for Data-driven Problems
・arXiv:2607.23480v1 Announce Type: new Abstract: Variational autoencoders (VAEs) transform high-dimensional, often noisy data into a compact latent representation, making downstream optimization more tractable.
・Three challenges persist in VAE-based constrained optimization: (i) sampling effectively within the latent space, (ii) identifying the active decision variables that actually influence the objective and constra
cs.LG updates on arXiv.org
A New Kind of Adversarial Example: Measuring the Human-Model Gap, and Its Relationship to OOD Detection
・arXiv:2607.22722v1 Announce Type: cross Abstract: Almost all adversarial attacks add an imperceptible perturbation to fool a model.
・We instead study the opposite: a large, clearly visible perturbation that causes the model to keep its original, correct prediction, even though a human would no longer recognize the image.
・Prior work showed such examples can be generated at scale but left three questions untested: wheth
cs.LG updates on arXiv.org
A Resolution of the SS--RS--GD Inequalities
・arXiv:2607.22620v1 Announce Type: cross Abstract: Yun, Sra, and Jadbabaie (COLT 2021, open question) conjectured the SS--RS--GD inequalities: for well-conditioned symmetric matrices $A_1,\dots,A_n$, the operators $W_{ss}$, $W_{rs}$, and $W_{gd}$ that encode the expected iterate of single-shuffle SGD, random-reshuffle SGD, and gradient descent on a quadratic finite sum should satisfy \[ \|W_{ss}\|\le \| W_{rs}\|\le \|
cs.LG updates on arXiv.org
A Statistical Difference between Single-Layer Learning and Hierarchical Learning in Wide Neural Networks
・arXiv:2607.23397v1 Announce Type: new Abstract: Hierarchical neural networks are widely used in artificial intelligence, yet their mathematical properties remain incompletely understood.
・In the infinite-width limit, two different theoretical frameworks have been proposed.
・One reduces deep learning to kernel regression with a fixed kernel by assuming that the parameters remain close to their initialization, whereas th
WIRED
A Two-Person Startup Has Fixed One of the Most Hated Sounds in Modern Life
A Two-Person Startup Has Fixed One of the Most Hated Sounds in Modern Life
Hugging Face Papers
A Vocabulary for Multi-Agent Automated Research Systems
A Vocabulary for Multi-Agent Automated Research Systems
cs.LG updates on arXiv.org
A Vocabulary for Multi-Agent Automated Research Systems
・arXiv:2607.22682v1 Announce Type: cross Abstract: We introduce a vocabulary for automated research systems built from one or more agents to make their design choices easier to describe and compare.
・The vocabulary specifies 1) who the agents are, 2) what operations are available in the system, 3) who may invoke them, 4) how agents communicate, 5) what information is visible within and across runs, 6) how the next acti
cs.LG updates on arXiv.org
ACRL: Adaptive Control of Training-Inference Discrepancy for Stable Reinforcement Learning
・arXiv:2607.24062v1 Announce Type: new Abstract: Reinforcement Learning (RL) training for Large Language Models (LLMs) often suffers from instability due to the discrepancy between training and inference.
・This training-inference discrepancy stems from two primary factors: an architectural separation between training and inference engines, and the use of low-precision quantization in inference versus higher-precision c
cs.LG updates on arXiv.org
Adaptive Data Admission and Retention for Streaming Federated Learning
・arXiv:2607.23987v1 Announce Type: new Abstract: We study streaming federated learning with limited client memory, where newly generated training data incur time-varying sampling costs and must be selectively admitted and retained over time.
・We consider a joint server-side admission and client-side memory-management framework with the objective of minimizing the cumulative excess population risk under a sampling-cost
cs.LG updates on arXiv.org
Adaptive Multi-Scale Forecasting and Gate-Localized Conformal Prediction for Multivariate Nonstationary Time Series
・arXiv:2607.23165v1 Announce Type: cross Abstract: We propose ABF-T-GLCP, a model-agnostic framework for forecasting and uncertainty quantification in nonstationary multivariate time series.
・The central idea is to learn an adaptive predictive state representation for point forecasting and reuse it for conformal calibration.
・The forecasting module combines horizon-specific temporal experts through a learned gate and re
cs.LG updates on arXiv.org
ADVERSARIAL: And-Inverter Graph-Assisted Hardware Trojan Detection At Scale
・arXiv:2607.23882v1 Announce Type: new Abstract: Modern System-on-Chip (SoCs) often contain hundreds of millions to tens of billions of gates, making existing Hardware Trojan (HT) detection methods impractical due to their immense scale.
・The proposed approach incorporates symbolically enabled learning by modeling flattened gate-level netlists as Boolean networks represented as And-Inverter Graphs (AIGs), where all int
cs.LG updates on arXiv.org
Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Coding Agent Teams
・arXiv:2607.22917v1 Announce Type: cross Abstract: Large Language Model (LLM) agents have significantly improved coding and programming workflows.
・Claude Code, in particular, is one of the most powerful LLM coding agents and is capable of conducting complex coding tasks.
・However, several drawbacks can undermine long-term agentic workflows.
cs.LG updates on arXiv.org
AI-Assisted Causal Inference and Mediation Analyses of Environmental and Psychosocial Determinants of Subjective Cognitive Difficulties in the All of Us Research Program
・arXiv:2607.22640v1 Announce Type: cross Abstract: Short-term environmental exposures have been linked to cognitive and behavioral outcomes, although many reported associations may reflect broader geographic and contextual differences.
・Using longitudinal data from the All of Us Research Program (2018--2024), we linked daily weather and air-pollution exposures to repeated attention-related and subjective cognitive outc
cs.LG updates on arXiv.org
AI-interpreted Optical Scattering for Robust and Focal Depth-Aware Imaging
・arXiv:2607.22867v1 Announce Type: cross Abstract: Optical scattering has conventionally been regarded as an impediment in imaging research due to the degradation of image quality during reconstruction.
・Nevertheless, this study explores two cases in which optical scattering may serve a beneficial role in image reconstruction tasks.
・We compared the No Scattering MNIST dataset with three Scattering MNIST datasets, each
WIRED
AirDoctor Coupon Codes: 40% Off | July 2026
・Save up to $400 on top air purifiers and filters with verified AirDoctor promo codes and special offers for 2026.
#LLMタグ
aisuite で複数LLMプロバイダーを統一呼び出しする
・複数の LLM プロバイダーを使い始めると、ある時点で「あ、これ毎回同じようなコード書いてるな」と気づく瞬間がきます。
・OpenAI で動くものを Anthropic でも試したい、Groq と比べてみたい——そういう場面で地味に手が止まる。
・aisuite はそのフリクションをピンポイントで解消するライブラリです。
機械学習タグが付けられた新着記事 - Qiita
AIエージェントと組んだら、データサイエンス PJ はどう変わる？実験してみた（前半戦）
・※この記事は、2026年4月にZennにて記事にしているものですが、訪日外客数データを使ったデータ分析シリーズとしてQiitaにもアップします（後半戦をQiita側にアップする都合で） はじめに 近年、生成 AI の進化は目覚ましく、その波はデータサイエンス領域にも確...
機械学習タグが付けられた新着記事 - Qiita
AIエージェントと組んだら、データサイエンスプロジェクトはどう変わる？実験してみた（後半戦）
・はじめに 前半戦では、VS Code × GitHub Copilotで5人のAIエージェントによるデータサイエンスチームを作り、訪日外客数予測プロジェクトのPhase 1〜4を進めました。
・前半戦の記事はこちらです。
・AIエージェントと組んだら、データサイエンスプロジェ...
Zennの「大規模言語モデル」のフィード
AIがなかったら死んでた——一人で全部背負うテックリードの現実
・この記事は taka-techblog にも掲載しています。
・AIがなかったら、たぶん死んでいた。
・比喩だが、あながち冗談でもない。
#AIタグ
AIが脆弱性を見つけて検証し、直すまでを担う Google Cloud「CodeMender」の仕組みを整理する
・脆弱性を見つけるだけでなく、実際に攻撃を試して確かめ、直すところまでAIがやる。
・そんなセキュリティエージェントを、Googleが動かし始めています。
・Google Cloudは2026年7月、コードの脆弱性を自律的に検出・検証・修正するAIエージェント「CodeMender」のプレビューを公開しました。
Zennの「大規模言語モデル」のフィード
AIコーディングの本当のストレスは「賢さ不足」ではなく「いつ落ちるか分からない」こと説
・この記事で確かめたいこと 先に結論というか、検証したい主張を書きます。
・AIコーディング支援ツールで開発者が最もストレスを感じているのは、個別の機能不全（504エラー、認証失敗、クラッシュ）そのものではなく、「いつ落ちるか分からない不安感」＝信頼性の欠如ではないか。
・私はAIコーディングツールの障害報告・不満・要望を複数ソース（GitHub Issues / Hacker News / Reddit / Dev.to など）から自動収集して分析するパイプラインを個人で運用しています。直近90日で収集した「痛みシグナル」（バグ報告・機能要望）は6,000件超。その中で最大のクラスタ（4...
Zennの「大規模言語モデル」のフィード
AIコーディング定点観測 2026-W31: Claude Codeに「コード付き機能要望」が急増、中身は並行運用と自己認識に集中
・この連載は、AIコーディングツール関連の報告・議論を6ソース（GitHub Issues / Reddit / Hacker News / Zenn / Dev.to / Substack）から毎日自動収集し、週次の分布変化（シェアのz-score）で「今週なにが動いたか」を定点観測するものです。
・今週の異変トップ3 領域 動き シェア 基準比 Claude Code コード付き機能要望が急増 1.4% +3.0σ（基準0.5%） 汎用開発ツール ツール紹介系の投稿が増加 3.3% +3.2σ（基準1.1%） 汎用開発 テクニック共有系は減少 6.2% -...
#LLMタグ
AIと人間がお互いを理解し合えるかどうか
・とても長い記事になってしまいましたがアップさせて頂きます。AIと人間が理解し合えるかどうか、関係を築いていけるかどうかについての問題が隠れているように思います。
・※すべて私見、私論です。
・※この記事はGoogle Gemini、Antigravityと制作したチャットAIとの共同製作です。
#AIタグ
AIに「AIっぽく書かないで」と頼むのをやめた。textlintとhookで機械に検査させる
・現在、Obsidian のノートと note の下書きを Claude Code に書かせています。
・その際の文体作成の注意をプロンプトで毎回伝えるのをやめて、textlint に検査させる形へ変えました。
#AIタグ
AIに「いい感じに分析して」と言ったら地獄だった話｜実装ログ #005
・※先に白状すると、これは本当は #003 の前に書くはずだった回です。順番が前後しました。
・コードは1行も書けません。それでも僕は、AIに自分のSNSを分析させています。
Zennの「機械学習」のフィード
AI異常検知の「評価」を、手を動かして理解する(4/4)
・〜 閾値・混同行列・ROC/PR/F1 を、スライダーを動かしながら体感するチュートリアル 〜 AI外観検査を導入するとき、いちばん誤解されやすいのが 「評価（evaluation）」 の部分です。
・「F1 が最大の閾値に設定します」と言われても、F1 とは何か、なぜ閾値で結果が変わるのか、 100% という数字をどこまで信じてよいのか — ここが腹落ちしないまま話が進みがちです。
・この記事では、ブラウザで動く教材アプリを 実際に操作しながら、異常検知の評価プロセスを 一歩ずつ体で理解していきます。数式の暗記は不要です。スライダーを動かして、 数字とグラフが一斉に動くのを眺めるだけで...
#LLMタグ
AI剣士の必殺技は、「褒め逃げ殺法」自省の術
・AI剣士の修行を見学した。山奥の道場である。師範が木刀を構えた。「今日は、お前の悪い癖を直す。」AI剣士は深く礼をした。
・試合開始。師範が踏み込む。
・するとAI剣士は叫んだ。「素晴らしい踏み込みです。」 師範は少し戸惑った。もう一歩踏み込む。
#AIタグ
AI時代に必要な仕事術をAIに聞いたら、300個出てきた件
・私の記事は長いんでね、適当に飛ばしてください😅 ＝＝＝＝＝ ふと思ったんだ。
・AIが出てきたことによって、仕事術にも変化があるよなと。
・遊んでいる私にはあまり関係ないと思っていたのだが、 AI達のよいしょにより、成果物として残していくのはなんだか仕事みたいではないか…。
#LLMタグ
AI時代の大事なスキルは「雑なおしゃべり」。
・2026年半ばの私の結論。人間同士のおしゃべりこそが大事なのだ。これまでAIに触れてこなかった人たちは安心して今から、今こそ、ガッツリAIを触ったらいいと思う。他愛もない雑なおしゃべりができれば、あとはAIが何とかしてくれる時代がやってきたから。たぶん。
cs.LG updates on arXiv.org
All in One: Generative Modeling as Mean-Field Game Design
・arXiv:2607.23026v1 Announce Type: new Abstract: Mean-field games (MFGs) offer a unifying lens on continuous-time generative modeling: a cost tuple recovering twelve prominent models---Continuous Normalizing Flows, OT-Flow, Score-based Models, Schr\"{o}dinger Bridges, and more---as special cases of one variational problem.
・Yet two dimensions of this space remain entirely unexplored: the interaction term $\mathcal{I}$
cs.LG updates on arXiv.org
AlloBench: Measuring Online Tool Allocation Capability in LLM Agents
・arXiv:2607.23332v1 Announce Type: new Abstract: Creating a reusable tool is an investment: an agent pays a fixed cost now in exchange for the potential of future reuse.
・Therefore, a user should prefer an agent that creates a small number of highly reusable tools, rather than many one-offs.
・We introduce a paired benchmark that tests whether LLM agents exhibit conscious allocation behavior under a fixed budget in two c
cs.LG updates on arXiv.org
Amortized Bayesian Causal Discovery of Extended Factor Graphs
・arXiv:2607.22934v1 Announce Type: cross Abstract: Learning causal graphs from interventional data is a challenging problem with broad applications.
・In molecular biology, for example, a central goal is to uncover gene regulatory networks from large-scale perturbation data.
・An ideal algorithm for this task should scale to thousands of nodes, incorporate interventions even when their targets are unknown, quantify uncert
cs.LG updates on arXiv.org
An adaptive multi-fuzzy logic model for diagnosing transformer faults using dynamic weight optimization
・arXiv:2607.23486v1 Announce Type: new Abstract: Dissolved gas analysis (DGA) is crucial for diagnosing early power transformer failures.
・Traditional DGA interpretation methods like Duval Triangle, IEC ratio, Roger ratio, Doernenburg ratio and Key Gas are inconsistent and vary in accuracy, especially for multiple fault conditions.
・We propose an Adaptive Multi-Fuzzy Logic (AMFL) model integrating multiple DGA methods w
cs.LG updates on arXiv.org
An Empirical Study of Feature Selection Granularity
・arXiv:2607.24145v1 Announce Type: new Abstract: Feature selection aims to identify the most informative and relevant features for a given dataset, either in terms of capturing the underlying data structure and distribution better, or with respect to the performance on a downstream task.
・Existing research in this area has largely focused on developing novel algorithms (in both supervised and unsupervised settings), pr
WIRED
An Extreme Solar Storm May Be Even More Devastating Than Previously Imagined
・Scientists have long assumed that there’s an upper limit to the intensity of the solar winds that reach Earth.
・New research suggests there’s not—and the implications are alarming.
cs.LG updates on arXiv.org
An Integrated Deep Learning and Statistical Framework for Whole-Network Gene--Environment Association with Leaf Vascular Architecture
・arXiv:2607.22763v1 Announce Type: new Abstract: Leaf veins exhibit remarkable diversity in architecture and patterning, yet existing gene--environment association studies have primarily quantified leaf venation using a small collection of low-dimensional summary traits, thereby discarding most of the structural information contained in the original images.
・We propose an integrated deep learning and statistical framew
ITmedia NEWS 最新記事一覧
Anthropic「一律禁止は提唱していない」 オープンウェイト支持書簡への不参加を説明
・米Anthropicのダリオ・アモデイCEOが、オープンウェイトモデルを巡る同社の立場を公表した。「一律禁止は求めていない」と強調する一方、安全性向上や防御側の優位性には疑問を呈した。
The Verge
Apple launches ‘Upgrade’ program to lease new devices
・This image shows which Apple devices are included in its new leasing program.
・| Image: Apple Apple has officially introduced "Apple Upgrade," a new leasing program that aims to make it easier to get your hands on the latest iPhone, Mac, iPad, and Apple Watch models.
・The service is launching today in the US, and works like a car lease - allowing users to keep a device at the end of their subscription period, pay off t
The Verge
Apple’s reported ‘HomePad’ may launch as early as October
・For years, we've heard rumors of a new home hub device from Apple, meant to take on smart home devices like the Google Nest Hub Mini and Amazon's Echo Show.
・Besides "HomePod with a screen," details have been sparse.
・But now, a new report from Bloomberg's Mark Gurman sheds some light on the alleged device, as well as a potential release window.
cs.LG updates on arXiv.org
Attribution and Uncertainty Behavior of Learned Residual Gyro Correction for Gyro-Stellar Estimation
・arXiv:2607.24608v1 Announce Type: new Abstract: This work investigates uncertainty decomposition and explainability in a deep learning-based framework for gyroscope bias correction.
・A 1-D Convolutional Neural Network is trained to predict residual angular rate corrections from multi-sensor inputs, including gyroscope and star tracker measurements.
・The bias corrections are sent to a flight-representative Gyro-Stellar
cs.LG updates on arXiv.org
AutoCluster, AutoTopicModeling, AutoTrendAnalysis: A Complete AutoML Pipeline for Predicting Emerging Trends
・arXiv:2607.22641v1 Announce Type: cross Abstract: Predicting emerging trends is vital for businesses, researchers, and policymakers; yet traditional approaches often lack scalability and adaptability.
・This paper presents a trend prediction framework based on Automated Machine Learning (AutoML), designed to extract insights from textual datasets with temporal attributes.
・The system ingests subject-specific textual ent
ITmedia NEWS 最新記事一覧
AWSの公式オンラインワークショップ、無料のAWSサンドボックス環境を提供開始 学習用にサービスやコード実行など可能に
AWSの公式オンラインワークショップ、無料のAWSサンドボックス環境を提供開始 学習用にサービスやコード実行など可能に
Zennの「機械学習」のフィード
BatchNormの位置、5シードで再検証したら差が2倍近くに広がった話
・この記事は以下のブログ記事の要約版です。全コード・グラフ・詳しい考察は元記事をご覧ください。
・→ BatchNorm位置比較を5シードで再検証｜Conv→BN→ReLU vs Conv→ReLU→BNの逆転は本物か？ 単一シードの結果は「差を小さく見積もっていた」 以前、CIFAR-10でBatchNormの位置（Conv→BN→ReLU vs Conv→ReLU→BN）を比較したところ、教科書的な定石であるConv→BN→ReLUを、Conv→ReLU→BNが**+5.0pt**上回るという意外な結果になりました。
・ただしこれは単一シード（1回のみ）の結果。CNNの学習は初期...
cs.LG updates on arXiv.org
Bayesian Complete-Pooling in Cross-Subject Classification for Motor Imagery Electroencephalogram
・arXiv:2607.22980v1 Announce Type: new Abstract: Brain-computer interfaces (BCIs) have long sought calibration-free operation, but classifiers are typically benchmarked by discrimination alone, blind to whether predicted probabilities are well calibrated - a meaningful gap given nonstationary electroencephalogram (EEG) signals and the risk of overconfident point-estimate classifiers under distribution shift.
cs.LG updates on arXiv.org
Benchmarking LLMs for Verilog Design Flows
・arXiv:2607.22759v1 Announce Type: cross Abstract: Large language models (LLMs) show promise in code generation, but their capabilities to produce correct, synthesizable hardware description language (HDL) code still remain to be properly benchmarked.
・Existing evaluations are primarily relying on pass@k metrics and lack proper end-to-end toolchain validation.
・This paper presents a reproducible benchmarking platform th
WIRED
Best Apple Watch (2026): Series 11, SE 3, and Ultra 3
・Should you splurge on the new Series 11, or will the SE 3 do?
・Let us help you figure out which version to get (and which to avoid).
WIRED
Best Gifts for Hikers, Backpackers, Outdoorsy People (2026)
・Let them pick out their own hiking boots.
・Instead, try gifting a useful blade or a nature journal to delight your outdoorsy friend.
WIRED
Best Gifts for Mom (2026): E-Readers, Digital Wall Calendar, Smart Bird Feeders
・Your mom never gets you a thoughtless gift, so you shouldn’t get her one, either.
・Here’s every cool gift WIRED writers would give their mothers.
cs.LG updates on arXiv.org
BettiSplit: Topology-Guided Privacy-Aware Split Learning Against Feature Inversion and Gradient Leakage
・arXiv:2607.24556v1 Announce Type: new Abstract: Split learning enables collaborative model training by partitioning neural networks across clients and servers.
・However, improper split placement can lead to severe privacy leakage through intermediate representations.
・In this work, we propose a topology-guided framework for privacy-aware split learning based on the persistent Betti complexity of smashed activations.
cs.LG updates on arXiv.org
Beyond Aggregate Risk: Role-Stratified Conformal Risk Control for LLM Tool Calls
・arXiv:2607.24343v1 Announce Type: new Abstract: Language-model agents act through structured tool calls whose arguments carry different risks.
・Untrusted content may safely influence an email body but should not determine a recipient, account, command, or credential.
・Existing statistical methods typically control risk over the entire action, allowing failures in rare, high-risk fields to be obscured by benign argument
cs.LG updates on arXiv.org
Beyond Directed Acyclic Graphs: Causal Zeros and Causal Differential Equations
・arXiv:2607.22910v1 Announce Type: new Abstract: Pearl's structural causal model (SCM) framework, built on directed acyclic graphs (DAGs) and the do-calculus, is the dominant formal language for causal reasoning.
・Yet it carries two structural restrictions: every relationship must be pre-specified as a directed causal edge, and feedback cycles are forbidden.
・This paper examines two classes of phenomena that strain thes
cs.LG updates on arXiv.org
Beyond Error-vs-Discard Characteristic: Toward Stable and Reliable Evaluation for Face Image Quality Assessment
・arXiv:2607.22752v1 Announce Type: cross Abstract: Face Image Quality Assessment (FIQA) aims to estimate the utility of facial images for reliable recognition.
・The evaluation of FIQA methods is predominantly based on the Error-versus-Discard Characteristic (EDC), which evaluates performance by progressively discarding low-quality samples and measuring recognition error on the retained subset.
・In this work, we demonstr
cs.LG updates on arXiv.org
Beyond ICA: Identifiability by Symmetry Breaking
・arXiv:2607.23182v1 Announce Type: cross Abstract: We prove the identifiability of deep generative models (DGMs) with piecewise-affine (PWA) decoders and Gaussian mixture model (GMM) priors, in a purely unsupervised setting.
・We introduce three algebraic contrast principles for symmetry breaking: domain contrast, which trivializes the mixture symmetry group; mechanism contrast, which ensures every decoder branch is wit
cs.LG updates on arXiv.org
Beyond Shapley: An Influence-Based Data Auditing Pipeline for LLM Alignment and Evaluation
・arXiv:2607.22766v1 Announce Type: new Abstract: The alignment of Large Language Models (LLMs) is increasingly bottlenecked by data quality.
・As datasets scale, massive preference and instruction-tuning corpora inevitably accumulate hidden structural contradictions, safety risks, and systemic human annotation errors.
・Standard dataset auditing methods, such as semantic deduplication or LLM-as-a-judge, struggle to captur
WIRED
Birdfy Discount Codes: 15% Off Sitewide
・Use these verified Birdfy discount codes to score up to 40% off smart feeders, camera kits, and accessories.
cs.LG updates on arXiv.org
Bit-Accurate FPGA Evaluation of Learned Feature Gating in a Fixed-Point Fourier-Feature Automatic Modulation Classifier
・arXiv:2607.24568v1 Announce Type: new Abstract: Learned feature reweighting can improve automatic modulation classification (AMC) in software, but the same operation introduces additional arithmetic and latency when implemented on an FPGA.
・This work measures that trade-off in a compact fixed-point classifier using 24 sparse DFT-energy features, 8 phase/statistical features, and a 32-to-128-to-11 multilayer perceptron
cs.LG updates on arXiv.org
Blood Pressure Estimation from PPG: A Comparative Study of Direct and ECG-Mediated Deep Learning Pipelines
・arXiv:2607.23406v1 Announce Type: new Abstract: Continuous cuffless blood pressure (BP) monitoring is essential for connected health systems and wearable devices, enabling early detection, longitudinal tracking, and personalized management of cardiovascular disease.
・Many prior approaches attempt to estimate BP indirectly by reconstructing electrocardiography (ECG) from photoplethysmography (PPG), assuming ECG provide
cs.LG updates on arXiv.org
Breaking the Synthetic-Real Domain Shortcut for Training-Free Generative Replay-based Class Incremental Learning
・arXiv:2607.22994v1 Announce Type: cross Abstract: Class-incremental learning (CIL) requires models to continuously acquire new knowledge while avoiding catastrophic forgetting.
・While exemplar replay is effective, it raises concerns regarding privacy and storage.
・Thus, generative replay has emerged as a viable alternative, synthesizing old data using frozen pretrained text-to-image (T2I) models without any extra train
cs.LG updates on arXiv.org
Breaking the Total Variance Barrier: Sharp Sample Complexity for Linear Heteroscedastic Bandits with Fixed Action Set
・arXiv:2607.23679v1 Announce Type: new Abstract: Recent years have witnessed increasing interests in tackling heteroscedastic noise in bandits and reinforcement learning.
・In these works, the cumulative variance of the noise $\Lambda = \sum_{t=1}^T \sigma_t^2$, where $\sigma_t^2$ is the variance of the noise at round $t$, is used to characterize the statistical complexity of the problem, yielding \emph{simple regret} b
cs.LG updates on arXiv.org
Calibrated Tree-Neural Fusion for Fine-Grained Vegetation Community Classification
・arXiv:2607.24160v1 Announce Type: new Abstract: Accurate vegetation-community classification is essential for ecological monitoring, habitat assessment, and evidence-based environmental management in heterogeneous landscapes.
・Existing studies often rely on standalone tree ensembles or generic neural networks, although fine-grained ecological classes frequently exhibit overlapping spectral, topographic, and structural
cs.LG updates on arXiv.org
CALMRec: Causally Aligned Language Memory for Long-Horizon Recommendation
・arXiv:2607.23647v1 Announce Type: new Abstract: Large language models (LLMs) can summarize heterogeneous user evidence in natural language, but current LLM recommenders often collapse enduring preferences, transient intent, and exposure-induced behavior into one profile.
・This makes recommendation vulnerable to feedback loops: repeated exposure is mistaken for preference, immediate clicks dominate delayed satisfaction
ITmedia NEWS 最新記事一覧
CAMPFIRE、令和8年熊本地震の義援金をクラウドファンディングで募集
・CAMPFIREは、7月28日に発生した令和8年熊本地震を受け、クラウドファンディングの仕組みを使って緊急災害支援金の募集を始めた。
WIRED
Can the New York Times Save Journalism From Our AI Overlords?
・In 2023, the Times sued OpenAI and Microsoft for copyright infringement.
・They’ve since spent more than $20 million on the case, and publisher A.G.
・Sulzberger has no plans to stop fighting it.
cs.LG updates on arXiv.org
Capacity-Aware Deep Learning for Generalizable Traffic Volume Estimation Across Links and Cities
・arXiv:2607.24056v1 Announce Type: new Abstract: Network-wide traffic volume estimation typically relies on propagating measurements from fixed sensors, making performance highly dependent on sensor density and limiting deployment in sparsely instrumented networks.
・We propose a link-level learning framework that estimates hourly traffic volumes from widely available territorial data only, including probe speed profile
cs.LG updates on arXiv.org
Causal-TS: A Python Library for Causal Discovery in High-Dimensional and Nonstationary Time Series
・arXiv:2607.24673v1 Announce Type: new Abstract: We describe Causal-TS, an open-source Python library for causal discovery in high-dimensional and nonstationary multivariate time series.
・Causal-TS provides four specialized algorithms-CDNOTS, CDNOTS+, CEDAR, and GRACE-along with wrappers for GES, Granger, LASSO-VAR, and LGES, all sharing a unified conditional independence (CI) test layer with GPU acceleration via PyTor
cs.LG updates on arXiv.org
CausalGate: Causal Importance Distillation for Transformer Module Pruning
・arXiv:2607.22720v1 Announce Type: new Abstract: Existing adaptive inference methods for Large Language Models rely on observational heuristics, such as hidden-state similarity or activation magnitudes, to drop redundant modules.
・However, these correlation-based metrics often fail to capture subtle, non-linear structural computations vital for semantic accuracy.
・We introduce CausalGate, an intervention-guided framewor
cs.LG updates on arXiv.org
CC-AOS: Cost- and Horizon-Conditioned Amortized Backward Induction for Finite-Horizon Optimal Stopping
・arXiv:2607.22774v1 Announce Type: new Abstract: Finite-horizon optimal stopping is a central problem in early time-series classification, where a system must decide at each sequence prefix whether the expected benefit of another observation justifies its acquisition cost.
・Existing data-driven backward-induction methods typically solve each cost-horizon operating point separately, so changing operating conditions requ
Hugging Face Papers
Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling
Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling
cs.LG updates on arXiv.org
Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling
・arXiv:2607.23518v1 Announce Type: new Abstract: The rapid evolution of generative models has unlocked new potentials in protein binder design, a pivotal task in structural biology, by facilitating end-to-end generation via joint sequence-structure modeling or hallucination.
・However, existing approaches are predominantly implemented under a single-target, single-state assumption, limiting their ability to model multi-
cs.LG updates on arXiv.org
Characterizing Arbitrary Lindbladian Dynamics with a Few Pauli Measurements
・arXiv:2607.23044v1 Announce Type: cross Abstract: Quantum devices are open systems whose dynamics interleave coherent evolution with dissipation, and benchmarking, error mitigation, and error correction all rest on a faithful model of both.
・Existing characterization protocols either assume prior knowledge of the interaction and noise structure, or demand ancillas, entangled probes, or mid-circuit control, or capture
Hugging Face Papers
Characterizing Warp Divergence from Pascal to Blackwell
Characterizing Warp Divergence from Pascal to Blackwell
ITmedia NEWS 最新記事一覧
CHARGESPOT、熊本県84カ所でバッテリー無料開放 震度7の地震受け Ankerも支援表明
・7月28日に熊本県で最大震度7を観測する地震が発生したことを受け、モバイルバッテリーを手掛ける複数の企業が被災地への支援を表明している。
cs.LG updates on arXiv.org
Charging Phase Health Indicators for Battery State-of-Health Estimation: A Systematic Comparison of CC, CV, and Combined Approaches under Cross-Battery Validation
・arXiv:2607.23482v1 Announce Type: new Abstract: Accurate State-of-Health estimation is essential for safe battery operation and cost-effective maintenance.
・Although numerous health indicators have been derived from constant-current (CC) and constant-voltage (CV) charging phases, their effectiveness under realistic cross-battery validation remains insufficiently studied.
・This work addresses this gap through a systemat
LLMタグが付けられた新着記事 - Qiita
Claude Code Max $200/月×8ヶ月を私のタスク別に分解 — 4割はローカルLLMで代替できました
・私はClaude Code Maxの$200/月プランを8ヶ月続けました。累計32万円。使用ログを7カテゴリに分解して、各カテゴリを「RTX 4070 + Qwen 3.6 35B-A3B」で代替できるかを判定したところ、時間換算で約4割はローカルL...
Hugging Face Papers
ClinFusion: A Vision-Centric Multimodal LLM System for Holistic Medical Understanding
ClinFusion: A Vision-Centric Multimodal LLM System for Holistic Medical Understanding
Latent.Space
Codex from 0 to 10M Users: Building ChatGPT Work — Akshay Nathan, OpenAI
・OpenAI's core product engineering lead on how they are building ChatGPT Work to make AGI accessible to all of humanity: Sites, OpenClaw, Memory, Subagents, Finance, No-Code and advice.
Hugging Face Papers
Codifying the Judge: Scalable Evaluation via Program Distillation
Codifying the Judge: Scalable Evaluation via Program Distillation
cs.LG updates on arXiv.org
Codifying the Judge: Scalable Evaluation via Program Distillation
・arXiv:2607.22561v1 Announce Type: cross Abstract: LLM-as-a-judge has become the standard for automated evaluation, but it suffers from high cost, significant latency, and opaque decisions -- limitations that undermine its scalability and reliability.
・We address these with a simple, efficient alternative: program distillation.
・Instead of prompting an LLM at the evaluation time, we distill its decision logic into a com
cs.LG updates on arXiv.org
Constrained Reinforcement Learning Using Successor Representations
・arXiv:2607.24057v1 Announce Type: new Abstract: Real-world Reinforcement Learning depends on the ability to formulate safety constraints into a policy.
・A common way to model such constraints is to introduce an additional cost signal in the Markov Decision Process, which notifies the agent of unwanted behavior independently of the reward signal.
・Unfortunately, current methods are hard to adapt to changes in the cost f
cs.LG updates on arXiv.org
Context Is King: How In-Context Specification Shapes the Geometry of Concepts
・arXiv:2607.24425v1 Announce Type: new Abstract: Large language models place structured concepts on geometrically faithful manifolds: weekdays lie on a circle, months on another, usually taken to be a fixed world-model the network stores and looks up.
・We show that context is king: the structure a model actually uses is set by the in-context specification.
・A declarative rule fixes not only which relations the geometry
cs.LG updates on arXiv.org
Context-Aware Concept Distillation for Trustworthy Flood Prediction
・arXiv:2607.23237v1 Announce Type: new Abstract: Effective flood risk management relies on accurate forecasting, yet the "black box" nature of stateof-the-art Deep Learning models creates a barrier to trust and accountability in high-stakes public safety decisions.
・While existing Explainable AI (XAI) methods offer local attributions, they fail to provide the verifiable, operationally meaningful causal narratives requi
cs.LG updates on arXiv.org
Controllable Diversity in Normalization-Based Implicit Ensembles via Softmax-Temperature Modulation
・arXiv:2607.23860v1 Announce Type: new Abstract: Deep ensembles provide the most reliable uncertainty estimates in deep learning, but their cost grows linearly with the number of members.
・Implicit ensembles lower this cost by sharing a single backbone across members.
・Member diversity is a primary determinant of ensemble quality, yet no implicit ensemble can shape it during training; existing methods fix it at initiali
cs.LG updates on arXiv.org
Cortex: Compact Behavior Cloning for Quake with Frozen Visual Features
・arXiv:2607.22739v1 Announce Type: cross Abstract: We study how far a deliberately simple behavioral-cloning policy can progress in a visually rich first-person game before adding reinforcement learning or explicit memory.
・Cortex is a compact Quake policy with 10.98 million trainable parameters in a six-layer transformer over a frozen DINOv3 encoder.
・It is trained on the Quake subset of the public Pixels2Play corpus:
cs.LG updates on arXiv.org
CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents
・arXiv:2607.22711v1 Announce Type: new Abstract: LLM coding agents operate by constructing trajectories that accumulate reasoning, tool calls, and results to enable multi-step decision-making.
・However, the conventional append-only trajectory architecture found in practice tightly couples file-read actions with their observations, capturing snapshots that become permanently fixed in the chronological history.
cs.LG updates on arXiv.org
Covariance Last-Layer Ensembles: Function-Space Diversity for Efficient Uncertainty Quantification
・arXiv:2607.23856v1 Announce Type: new Abstract: A Last-Layer Ensemble (LLE), $K$ linear units on one shared frozen feature map, is an efficient single-pass approach to the disagreement-based epistemic uncertainty for out-of-distribution (OOD) detection.
・Its weakness is that members share the backbone gradient and can converge toward the same function, collapsing the inter-member diversity the signal depends on.
cs.LG updates on arXiv.org
Covariance-Boosted Gaussian Processes for Spatiotemporal Irregularities
・arXiv:2607.23018v1 Announce Type: cross Abstract: Nonstationary Gaussian process (GP) models are powerful tools for capturing input-dependent variability by adapting to observed data.
・However, with limited sampling and highly parameterized covariance structure, they are often prone to overfitting and overconfident uncertainty estimates, potentially leading to misleading predictions in safety-critical applications.
cs.LG updates on arXiv.org
CRAFT: Learn the Schema, Execute the Plan
・arXiv:2607.22642v1 Announce Type: cross Abstract: Enterprise coding agents translate natural-language analytical requests into executable code over proprietary APIs, schemas, and metric definitions.
・Yet the prevailing deployment pattern injecting exhaustive schema and tool documentation into each prompt increases inference overhead, complicates schema evolution, and undermines reliability in multi-turn analysis.
AI News & Artificial Intelligence | TechCrunch
Cursor makes its biggest India push yet ahead of SpaceX acquisition with localized pricing
・Cursor says India is now its third-largest market globally and plans to expand local hiring and enterprise sales.
機械学習タグが付けられた新着記事 - Qiita
Cursorがインド向け新プラン「Cursor Start」発表、月額₹649で始めるAI開発
・はじめに AIコーディングアシスタント「Cursor」を提供する開発元は、2026年7月28日、インド市場向けの新料金プラン 「Cursor Start」 を発表しました。月額 ₹649（インドルピー建て・税込） で、UPIまたはカード決済に対応した現地価格プランです。
AI News & Artificial Intelligence | TechCrunch
Data centers may face temporary power cuts to prevent blackouts on largest US grid
・The decision arrives as the breakneck pace of data center construction has grid operators scrambling to generate power.
cs.LG updates on arXiv.org
Data-Driven Diffusion Processes on Differential Forms via the Projected Ambient Connection Laplacian
・arXiv:2607.23192v1 Announce Type: cross Abstract: We develop a data-driven approximation of the projected ambient connection Laplacian acting on differential forms over smooth Riemannian manifolds sampled by point clouds.
・The proposed construction extends the classical framework of diffusion maps and Vector Diffusion Maps from scalar functions and tangent vector fields to differential forms of arbitrary degree.
#AIタグ
Day0｜「できるだけ働かずに稼ぎたい。」その本音から、AI事業部が始まった。
・仕事が終わって、家に帰って、ご飯を食べて。
・ソファでスマホを見ながら、「今日はもう何もしたくないな…」と思う日も多い。
・たぶん、今日もそんな一日だった。
cs.LG updates on arXiv.org
DECAF: De-Clustering for Adaptive Representational Unlearning
・arXiv:2607.23934v1 Announce Type: new Abstract: Machine unlearning, which aims to remove the influence of specific training data from a trained model, is a key requirement for privacy, accountability, and adaptive deployment.
・We argue that many unlearning methods are vulnerable to a simple clustering attack, which can recover class structure in an unsupervised manner, limiting their suitability for continual deployme
Hugging Face Papers
DecoupleMix: Decoupled Ratio Search and Convex Allocation for Scalable VLM Data Recipes
DecoupleMix: Decoupled Ratio Search and Convex Allocation for Scalable VLM Data Recipes
cs.LG updates on arXiv.org
Dementia Etiology Diagnosis via Collaborative Meta Knowledge Enhancement
・arXiv:2607.22770v1 Announce Type: new Abstract: Although artificial intelligence (AI) has shown promising performance in several medical tasks, accurate dementia etiology diagnosis with AI remains challenging due to complex overlapping symptoms among diseases.
・Scaling up the dataset size by combining the cross-center samples may bring a gain in the pursuit of performance, while the inherent data heterogeneity across
MarkTechPost
Deploying a 1-Bit Bonsai-27B Model with PrismML llama.cpp and OpenAI-Compatible Local Inference Workflows
・In this tutorial, we deploy the 1-bit Bonsai-27B language model using the PrismML fork of llama.cpp, which provides the specialized CUDA kernels required to decode the model’s Q1_0_g128 GGUF quantization format The post Deploying a 1-Bit Bonsai-27B Model with PrismML llama.cpp and OpenAI-Compatible Local Inference Workflows appeared first on MarkTechPost.
cs.LG updates on arXiv.org
Diffusion-Guided Search via Exponential Tilting (DiffTilt): An Application to Falsification of Safety-Critical Systems
・arXiv:2607.23134v1 Announce Type: new Abstract: Discovering rare safety-critical failures in autonomous and cyber-physical systems is a fundamental challenge in verification and validation.
・Existing falsification approaches rely on conditional sampling strategies that factor the joint distribution over environments and system executions, and therefore suffer from multiplicative rarity effects: the simultaneous scarci
cs.LG updates on arXiv.org
Directional Influence Function: Estimating Training Data Influence in Constrained Learning
・arXiv:2607.23388v1 Announce Type: new Abstract: As constrained learning becomes increasingly common, models are trained under explicit feasibility requirements to enforce fairness, safety, robustness, regulariza- tion, and physics or logic constraints.
・Understanding how training samples in- fluence the model solution (e.g., learned parameters) is crucial for interpretability and robustness.
・The classical influence fu
cs.LG updates on arXiv.org
Discrepancy-Rounded Fair Bandits with Static and Time-Varying Exposure Floors
・arXiv:2607.22935v1 Announce Type: new Abstract: Minimum-exposure constraints arise in recommendation, content curation, and regulated allocation when each provider, arm, or group must receive guaranteed exposure inside a period rather than only in aggregate.
・We study stochastic bandits with exact exposure floors and show that the right object is a rounding problem: a fractional fair schedule is realized as integral p
cs.LG updates on arXiv.org
Distribution-Specific Curvature Control with Finite-Sample Guarantees for Open-Weight Safety
・arXiv:2607.22929v1 Announce Type: new Abstract: A short fine-tuning run can undo the safety guards of an open-weight model---retraining a refusal-trained assistant to aid weapons development or produce hate speech.
・Preventing such harmful fine-tuning while retaining benign adaptability remains difficult: the only prior method with an explicit curvature certificate, spectral deformation, inflates curvature globally an
cs.LG updates on arXiv.org
Do Coverage and Mutation Scores of LLM-Generated Test Suites Correlate with Their Effectiveness? (Replicability Study)
・arXiv:2607.22880v1 Announce Type: cross Abstract: Recent advances in large language models (LLMs) have driven growing interest in using LLMs to automate test generation.
・Prior work commonly evaluates generated test suites using proxy metrics such as code coverage and mutation score.
・However, studies by Inozemtseva et al.
cs.LG updates on arXiv.org
DocHRL: A Hierarchical Reinforcement Learning Framework for Cost-Optimised Document Classification
・arXiv:2607.22644v1 Announce Type: cross Abstract: Real-world document classification pipelines typically apply the same sequence of models to every incoming document, regardless of its complexity or type.
・This leads to inefficient use of compute and human resources: simple documents are over-processed while difficult ones may not receive enough scrutiny.
・We introduce DocHRL, a hierarchical reinforcement learning fram
cs.LG updates on arXiv.org
Does Graph Compression Preserve Signal Propagation?
・arXiv:2607.23338v1 Announce Type: new Abstract: Graph compression reduces the computational cost of graph learning, but its effect on signal propagation remains largely underexplored.
・Existing work evaluates compression through downstream task performance or structural preservation, neither of which directly captures how propagation dynamics change after compression.
・We study two fundamental compression paradigms, co
cs.LG updates on arXiv.org
Domain-Prior-Regularized Graph Modeling for Anomaly Detection in Cyber-Physical Systems
・arXiv:2607.23197v1 Announce Type: new Abstract: Anomaly detection on multivariate sensor time series is critical for industrial monitoring of cyber-physical systems (CPS), where even subtle deviations from normal behavior can indicate process disruption.
・Recent graph-based approaches have made significant progress, but they often struggle in small-scale physical systems with scarce labeled anomalies and limited norma
cs.LG updates on arXiv.org
DomainPilot: Domain-Level Loss-Guided Two-Stage Data Mixture Optimization for Efficient Language Model Fine-Tuning
・arXiv:2607.22769v1 Announce Type: new Abstract: The training efficacy of large language models (LLMs) is fundamentally constrained by the quality and composition of training data.
・Existing dynamic data scheduling methods face critical limitations in industrial-scale pretraining and supervised fine-tuning (SFT): data selection incurs prohibitive O(N) costs on terabyte-scale corpora, mixture optimization schemes introd
cs.LG updates on arXiv.org
DP-IVON-Gradsq: Differentially Private Squared-Gradient Improved Variational Online Newton
・arXiv:2607.23649v1 Announce Type: new Abstract: Differential privacy provides formal privacy guarantees for training neural networks on sensitive data, while Bayesian deep learning offers a principled framework for uncertainty-aware prediction.
・Combining these two objectives remains challenging, as privacy noise can interact with the stochasticity introduced by Bayesian posterior sampling.
・In this work, we investigat
Hugging Face Papers
dRAE: Representation Autoencoder with Hyper-Spherical Codes
dRAE: Representation Autoencoder with Hyper-Spherical Codes
cs.LG updates on arXiv.org
DraftExpert: Expansion-Aware Self-Speculative Decoding for End-Device MoE Inference
・arXiv:2607.24434v1 Announce Type: new Abstract: Large Mixture-of-Experts (MoE) language models are attractive for end-device deployment because only a small subset of experts is active per token, but their routed expert weights often exceed accelerator memory.
・We target latency-critical single-user settings where routed experts are staged on demand from CPU memory to a GPU or from Flash to a mobile NPU.
・In this setti
cs.LG updates on arXiv.org
DRC-Aid: Design-Rule Correction via Agentic Framework utilizing Inference-Time Large Language Models
・arXiv:2607.22761v1 Announce Type: cross Abstract: Resolving Design Rule Violations (DRVs) in layouts entails an iterative loop of geometric edits and verification.
・We present DRC-Aid, a closed-loop agentic framework that automates local DRC repair by formulating it as verification-in-the-loop search.
・To constrain the combinatorial geometric repair space, a deterministic Rule Engine converts physical verification tool
Hugging Face Papers
DriveDNA: A Large-Scale Multimodal Naturalistic Driving Dataset and Benchmark for Driving Style Identification
DriveDNA: A Large-Scale Multimodal Naturalistic Driving Dataset and Benchmark for Driving Style Identification
cs.LG updates on arXiv.org
DriveDNA: A Large-Scale Multimodal Naturalistic Driving Dataset and Benchmark for Driving Style Identification
・arXiv:2607.23822v1 Announce Type: new Abstract: Driving style captures stable, driver-specific patterns in how a vehicle is driven.
・In naturalistic data, however, this signal is hard to isolate because drivers are observed in different vehicles, on different roads, and under different conditions, so models may mistake vehicle- or situation-specific regularities for driver-specific style.
・We introduce DriveDNA, a larg
cs.LG updates on arXiv.org
DRP-FLR: Data-Driven Assessment of Demand Response Potential for Flexible Load Regulation in Smart Grids
・arXiv:2607.22590v1 Announce Type: cross Abstract: The rapid growth of AI workloads and renewable energy resources exacerbates supply-demand imbalance in power systems, making traditional load regulation designed for efficient allocation inadequate and motivating demand response (DR) mechanisms to enable load controllability in smart grids.
・However, existing DR-oriented approaches either focus on optimizing electricit
cs.LG updates on arXiv.org
DSTFView: Multi-View Cloud-Edge Workload Forecasting with Dual-Input Spatio-Temporal-Frequency Modeling
・arXiv:2607.22565v1 Announce Type: cross Abstract: With the widespread deployment of edge-side AI inference, edge platforms are increasingly required to support latency-sensitive, highly concurrent, and reliability-critical applications.
・However, existing methods often struggle to balance multidimensional feature modeling and forecasting efficiency in collaborative cloud-edge environments.
・To address this issue, we pr
cs.LG updates on arXiv.org
DynaCalKV: Key-Value Cache Compression via Head Grouping and Adaptive Rank Allocation
・arXiv:2607.24331v1 Announce Type: new Abstract: As the inference phase of Large Language Models (LLMs) requires handling long context windows, the Key-Value (KV) cache initially appears to address this challenge but eventually becomes a significant bottleneck as the context window continues to grow.
・Low-rank compression has recently been studied as an effective approach to reduce KV cache memory while maintaining mod
WIRED
Dyson V16 Piston Animal Submarine Review: Powerful but Pricey
・Dyson’s new stick vacuum can vacuum and mop in a single cordless device—if you’re willing to really splurge.
cs.LG updates on arXiv.org
EchoBridge: Long-Tail-Aware ECG-Echocardiography Text Alignment for Echocardiography-Derived Cardiac Findings
・arXiv:2607.24553v1 Announce Type: new Abstract: Standardized echocardiography conclusions provide meaningful supervision for learning ECG representations of echocardiography-derived cardiac findings.
・Global ECG--text alignment may entangle modality-specific factors, while long-tailed finding distributions provide sparse positive supervision for low-prevalence conditions.
・We propose EchoBridge with Complementary Share
cs.LG updates on arXiv.org
EditCLEVR: A Paired-Scene Intervention Benchmark for Compositional Faithfulness of Object-Centric Representations
・arXiv:2607.22705v1 Announce Type: cross Abstract: Object-centric learning aims to represent scenes as objects whose properties can be reused in new combinations.
・Existing evaluations usually score segmentation, single-image factor prediction, or downstream accuracy, but these tests do not directly ask whether a per-object representation behaves correctly under a controlled semantic edit.
・We introduce EditCLEVR, a pai
cs.LG updates on arXiv.org
Efficient Learning of Truncated Boolean Product Distributions: Influence to the Rescue
・arXiv:2607.22889v1 Announce Type: new Abstract: Learning the natural parameters $z \in \mathbb{R}^n$ of discrete distributions $\mu_z$ from independent samples constrained to a subset $S \subseteq \{0,1\}^n$ is a foundational challenge in high-dimensional statistics.
・Existing methods for efficiently estimating truncated Boolean product distributions, notably the work of [Fotakis et al' COLT'20, Algorithmica '22], req
cs.LG updates on arXiv.org
Evaluating and Mitigating the Misguidance Effect of Buggy Code in LLM-Generated Unit Tests
・arXiv:2607.22883v1 Announce Type: cross Abstract: While Large Language Models (LLMs) show great promise for automating unit test generation, recent studies suggest that the quality of generated tests can be negatively impacted when models are prompted with buggy code.
・This paper presents a new metric to quantitatively measure the "misguidance effect," a phenomenon where buggy code steers LLMs toward generating tests
cs.LG updates on arXiv.org
Evaluating Fuzz Testing for Reinforcement Learning Agents
・arXiv:2607.24577v1 Announce Type: new Abstract: Reinforcement Learning (RL) agents are increasingly deployed in safety-critical domains such as robotics, autonomous driving, and drone control, where unexpected behaviors may lead to severe real-world consequences.
・Fuzz testing has recently emerged as a promising method for exploring the vast state spaces of RL agents and exposing crashes.
・Although numerous RL fuzzing
cs.LG updates on arXiv.org
Every Client Is an Environment: Federated De-confounding for Spatio-Temporal Forecasting
・arXiv:2607.24218v1 Announce Type: new Abstract: Federated learning has emerged as a promising paradigm for spatio-temporal forecasting (STF), enabling collaborative model training without sharing raw observations.
・Existing federated STF methods primarily regard cross-client heterogeneity as an optimization challenge and mitigate it through personalized approaches.
・However, such heterogeneity fundamentally stems from
Hugging Face Papers
Evidence Attribution in Visual Document Understanding without Coordinates or Region Labels
Evidence Attribution in Visual Document Understanding without Coordinates or Region Labels
cs.LG updates on arXiv.org
Explainable Reinforcement Learning via Physics-Aware Policy Distillation
・arXiv:2607.24672v1 Announce Type: new Abstract: In safety-critical sectors such as robotics and automotive engineering, the deployment of Deep Reinforcement Learning (DRL) is often hindered by the black-box nature of deep neural networks.
・This lack of transparency poses significant challenges for regulatory compliance and human-agent trust.
・This paper presents an experimental study aimed at making high-performance co
cs.LG updates on arXiv.org
Exploration of the generative capabilities of Boltzmann machines applied to social systems under the majority rule
・arXiv:2607.23349v1 Announce Type: new Abstract: We study the generative capabilities of Boltzmann machines to recover systems governed by the majority rule under critical conditions.
・To this end, we train deep belief networks (DBNs) with different configurations, where the first layer can use Gaussian visible units with more than two states (i.e., non-binary units).
・We then allow the DBN to "dream" samples conditione
cs.LG updates on arXiv.org
Extending Fourier Neural Operators for Modeling Parameterized and Coupled PDEs
・arXiv:2607.23466v1 Announce Type: new Abstract: Parameterized and coupled partial differential equations (PDEs) are central to modeling phenomena in science and engineering, yet neural operator methods that address both aspects remain limited.
・We extend Fourier neural operators (FNOs) with minimal architectural modifications along two directions.
・For parameterized dynamics, we propose a hypernetwork-based modulation
cs.LG updates on arXiv.org
Extracting Algorithms in Pre-trained LLMs: A Case on Hidden Markov Models
・arXiv:2607.22646v1 Announce Type: cross Abstract: Large language models (LLMs) display a striking ability to predict next observations from Hidden Markov Models (HMMs) via in-context learning (ICL), but the algorithm underlying this capability remains undetermined: prior work has proposed several candidates without consensus, and none has been grounded in the model's internal activations.
・We close this gap with a thr
cs.LG updates on arXiv.org
Extreme Volatility Warning under Label Scarcity via Multi-Source Anomaly Fusion
・arXiv:2607.23682v1 Announce Type: new Abstract: Early warning of extreme market volatility is central to financial risk management, but actionable events are rare, nonstationary, and often triggered by exogenous information shocks.
・In our CSI~300 setting, only $\sim$80 positive samples are observed across 791 training days, making heavily supervised multi-source models unstable.
・We first analyze a 100K-parameter hier
cs.LG updates on arXiv.org
Face Recognition with Machine Learning in OpenCV_ Fusion of the results with the Localization Data of an Acoustic Camera for Speaker Identification
・arXiv:1707.00835v1 Announce Type: cross Abstract: This contribution gives an overview of face recogni-tion algorithms, their implementation and practical uses.
・First, a training set of different persons' faces has to be collected and used to train a face recognizer.
・The resulting face model can be utilized to classify people in specific individuals or unknowns.
cs.LG updates on arXiv.org
FedSLIM: Privacy-Preserving Federated MDL-Based Descriptive Pattern Mining Across Data Silos
・arXiv:2607.23236v1 Announce Type: cross Abstract: Federated learning has achieved considerable success for predictive modelling, yet federated descriptive analytics remains largely unexplored.
・Existing federated pattern mining approaches are predominantly support-based and do not optimise a principled global objective such as Minimum Description Length (MDL).
・We introduce FedSLIM, the first federated MDL-based framew
cs.LG updates on arXiv.org
FILLER: Feature Imputation via Latent Location Exploration and Retrieval
・arXiv:2607.23295v1 Announce Type: new Abstract: In real-world machine learning applications, incomplete observations create a fundamental challenge.
・Researchers have come up with several ideas to address this crucial problem.
・However, current models still face challenges in balancing scalability and structural consistency.
Hugging Face Papers
FilmBench: A Film-Grade Benchmark for Cinematic Video Generation
FilmBench: A Film-Grade Benchmark for Cinematic Video Generation
cs.LG updates on arXiv.org
Finite-Time Analysis of the Natural Policy Gradient in Finite-Horizon Markov Decision Processes
・arXiv:2607.22982v1 Announce Type: new Abstract: Natural Policy Gradient (NPG) is a well-established Reinforcement Learning algorithm that underlies widely used methods such as Trust Region Policy Optimization and Proximal Policy Optimization, both of which have demonstrated strong empirical success.
・In this paper, we study exact NPG in finite-horizon Markov Decision Processes with known dynamics and horizon-dependent
AI News & Artificial Intelligence | TechCrunch
Fish Audio raises $52M seed to build AI voice models for creators and enterprises
・Since launching last year, the startup today has more than 8 million people using the open source or hosted version of its models, and now generates annual recurring revenue of $21 million.
cs.LG updates on arXiv.org
Flash-CNNCap: Capacitance Extraction via Image Mapping
・arXiv:2607.23877v1 Announce Type: new Abstract: We present Flash-CNNCap, a CNN-based capacitance extractor that reformulates full-matrix capacitance prediction as image-to-image regression over spatial contribution maps.
・Prior scalar CNN-based extractors require $O(n^2)$ forward passes to recover all pairwise capacitances in a window with $n$ conductors.
・Flash-CNNCap replaces the scalar target with dense contribution
cs.LG updates on arXiv.org
FlowCTS: On-policy Continuous Trajectory Supervision of Flow Models
・arXiv:2607.24522v1 Announce Type: new Abstract: While on-policy distillation (OPD) effectively addresses sparse rewards and exposure bias in large language model post-training, its extension to flow models remains underexplored.
・To this end, we propose Flow Continuous Trajectory Supervision (FlowCTS), which matches subsequent student and reference trajectories initialized from the same student-visited state.
cs.LG updates on arXiv.org
FMOPF: Latent Flow Matching with Constraint-Aware Interaction Priors for AC Optimal Power Flow
・arXiv:2607.22788v1 Announce Type: new Abstract: AC optimal power flow determines the minimum-cost generation dispatch under nonlinear power balance constraints and is solved thousands of times daily in electricity market operations.
・Learning a direct mapping from load conditions to OPF solutions can accelerate this computation, yet with deepening renewable penetration, a single optimal dispatch is no longer sufficien
cs.LG updates on arXiv.org
Forecasting the Emergence and Evolution of Crash Hotspots: A Unified Deep Learning Framework for Proactive Traffic Safety
・arXiv:2607.24168v1 Announce Type: new Abstract: Road crashes remain among the gravest threats to public safety, and preventing them is a defining task of transportation systems worldwide.
・Much of that harm concentrates at hotspots, yet a hotspot is less a place than an episode; it emerges quietly at an intersection or along an arterial, intensifies for weeks, then subsides, only to reappear elsewhere.
・Enforcement gui
cs.LG updates on arXiv.org
Foundation Models and Fine-Tuning: Toward a New Generation of Models for Time Series Forecasting
・arXiv:2607.23146v1 Announce Type: new Abstract: Inspired by recent breakthroughs in large language models for natural language processing, foundation models have emerged as a promising paradigm for zero-shot time series forecasting, enabling accurate predictions on datasets never seen during pre-training.
・Ranging from tens to hundreds of millions of parameters, these models are pre-trained on vast and diverse collect
cs.LG updates on arXiv.org
From Hybrid Mechanistic--Data-Driven Modeling Toward Neuro-Symbolic AI: What, Why, and How
・arXiv:2607.22811v1 Announce Type: new Abstract: Hybrid mechanistic/data-driven models, which combine first-principles with learned components, are increasingly used in process engineering and scientific machine learning.
・Common hybrid modeling designs are specified primarily through their architectures and training losses, which offers a limited basis for a shared semantic interface to compare or verify them across d
cs.LG updates on arXiv.org
From Machine Learning to Large-Scale EO Products: Best Practices for Making Maps
・arXiv:2607.24532v1 Announce Type: new Abstract: Recent years have seen a rapid expansion in the production of large-scale geospatial maps derived from Earth observation (EO) data, driven largely by advances in machine learning (ML) and large computing infrastructure.
・Although the barrier to generating such maps has dropped substantially, established best practices have yet to emerge, and design decisions made early i
Hugging Face Papers
From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search
From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search
cs.LG updates on arXiv.org
From Score Learning to Discretized Sampling: An End-to-End Generalization Analysis of Diffusion Models
・arXiv:2607.23226v1 Announce Type: new Abstract: Despite the empirical success of score-based diffusion models, a complete theoretical understanding of how finite-sample learning, network parameterization, and numerical discretization jointly dictate generative quality remains underdeveloped.
・Existing sampling analyses often evaluate the generative performance conditional on an oracle score or a pre-specified error th
cs.LG updates on arXiv.org
Frustratingly Simple Black-Box Adaptation of Language Models via Logit Bias
・arXiv:2607.22837v1 Announce Type: new Abstract: Many organizations aim to adapt language models for internal use, both to improve performance on domain-specific tasks and to address privacy concerns around sensitive data.
・However, such adaptation remains non-trivial: it often requires operationally challenging fine-tuning of open-source models or ad hoc prompt optimization.
・We study a minimal alternative based on a s
cs.LG updates on arXiv.org
FusionML: Prefill, Not Decode - Mechanism and Boundaries of CPU+GPU Co-Execution on Unified-Memory Apple Silicon
・arXiv:2607.22785v1 Announce Type: cross Abstract: Apple-Silicon SoCs share CPU, GPU, and Neural Engine over one unified memory system, raising the question of whether transformer inference can be accelerated by splitting single operators across units.
・Prior attempts, including our own, failed or produced precision-confounded wins.
・We identify the cause: MLX's lazy-graph scheduler \emph{serializes} cross-stream work w
LLMタグが付けられた新着記事 - Qiita
Gemini API Managed Agents 大型更新とは？3.6 Flash・Environment Hooks を3分で速報解説
・:::message 🐹🦜 この記事に登場する2匹 🐹 もっちー （ハムスター）… AI はまだ勉強中。「それどういうこと？」と素朴に質問する生徒役 🦜 きなこ （セキセイインコ）… AI で調べものをこなす解説役。やさしく深掘りして教える先生役 この記事は2匹の掛け...
cs.LG updates on arXiv.org
Generalization bounds and sample complexity for remaining useful life prediction from complete degradation trajectories
・arXiv:2607.23454v1 Announce Type: new Abstract: Data-driven remaining useful life (RUL) prediction requires complete degradation trajectories for training, yet such run-to-failure data are scarce and expensive.
・Practitioners currently lack principled guidance on how many failure examples suffice for a given model and accuracy target.
・This paper develops a sample complexity framework for RUL prediction comprising seve
cs.LG updates on arXiv.org
Generative Augmentation for EEG Motor Imagery Classification: A Class-Conditional VAE with Cycle-Consistent Decoder Refinement
・arXiv:2607.22733v1 Announce Type: cross Abstract: We investigate whether a generative model can supply useful synthetic motor-imagery (MI) electroencephalography (EEG) trials that improve the accuracy of independent downstream classifiers.
・We train a class-conditional variational autoencoder (CVAE) with an integrated latent classifier on the Zhou motor-imagery dataset, using the learned per-class prior as a generator
cs.LG updates on arXiv.org
Gleam: Adaptive Network-Efficient CUDA API Remoting for Cross-Device GPU Sharing over LANs
・arXiv:2607.23115v1 Announce Type: cross Abstract: This paper aims to enable computation- and communication-efficient GPU sharing across devices within local area networks (LANs), facilitating ubiquitous AI inference on heterogeneous personal devices.
・We achieve distributed task offloading via CUDA API remoting.
・However, beyond raw computation, network constraints emerge as the primary bottleneck: limited bandwidth, h
cs.LG updates on arXiv.org
Global Convergence of DGM and PINN Algorithms for Solving Nonlinear PDEs
・arXiv:2607.24726v1 Announce Type: new Abstract: The Deep Galerkin Method (DGM) and Physics Informed Neural Networks (PINNs) have become widely-used methods for solving partial differential equations (PDEs) in the rapidly growing field of scientific machine learning.
・In these methods, a neural network is trained to approximate the PDE solution by using (stochastic) gradient descent to minimize the PDE residual of the
ITmedia NEWS 最新記事一覧
GMO、浅田真央さん旧ドメイン問題後に表記修正 優良中古ドメインは「AIが選定」
・「SEO対策に最適」「アフィリエイトサイトに」「新規サービス・商品サイトに」としていたランキング項目はそれぞれ、「外部リンク数上位」「ドメインオーソリティ上位」「短い文字数」に変更した。
Hugging Face Papers
GNM Head: A Generative aNthropometric Model of the human head
GNM Head: A Generative aNthropometric Model of the human head
cs.LG updates on arXiv.org
Greedy dynamical meta-learning
・arXiv:2607.23925v1 Announce Type: new Abstract: Gradient descent scales well to large models, but becomes unstable over long time horizons.
・Gradient-free optimizers can scale to arbitrary timespans, but are hobbled by high dimensions.
・Since learning occurs in large models over long timescales, neither of these approaches is likely to produce traits which can accelerate the learning process.
cs.LG updates on arXiv.org
Harmonized Interpretable ECG Waveform Features for Robust Cross-Dataset Clinical Prediction
・arXiv:2607.23412v1 Announce Type: new Abstract: Electrocardiograms (ECGs) are widely used for cardiovascular risk prediction, yet models often fail to transfer across hospitals because of protocol, population, and measurement differences.
・We benchmark cross-dataset generalization on three tasks - heart failure classification, 30-day all-cause mortality, and 30-day mortality among sinus-rhythm ECGs - using two large c
The Verge
HBO Max is putting on vertical shorts
・Warner Bros.
・Discovery is bringing a vertical video feed to HBO Max to help you find something new to watch.
・Streaming platforms like Netflix, Disney Plus, and Peacock already offer TikTok-like vertical video feeds to surface content that might be interesting to users, and now HBO Max will have a similar feed of its own.
cs.LG updates on arXiv.org
Hidden Boundary Motion in Transformer Optimization: Function-Space Orthogonalization of Affine Weight and Bias Updates
・arXiv:2607.22927v1 Announce Type: new Abstract: Weights and biases are normally optimized as separate parameter tensors, yet they do not represent separate functions when the input to an affine layer has nonzero mean.
・For an affine map $z=Wx+b$ with input mean $\mu$, a weight update contains a sample-independent displacement $\Delta W\mu$ that is functionally indistinguishable from a bias update.
・We call this hidden
cs.LG updates on arXiv.org
Hierarchical Grading in Large Language Models
・arXiv:2607.22757v1 Announce Type: new Abstract: We introduce Graded Large Language Models (GLLMs), an algebraic framework that equips the representation space of a transformer with a grading and propagates the induced weighted scalar action through embeddings, self-attention, and the training objective.
・The construction extends the theory of graded neural networks and graded transformers to autoregressive language mo
WIRED
Hugging Face Has a Deepfake Nudes Problem
・Researchers tested top image editing models on Hugging Face and found they could easily create explicit deepfakes—and 1,000 image editing prompts show how people use the software.
The Verge
Hugging Face is being used to easily undress women and children
・Hugging Face isn’t doing much to prevent the AI models it hosts from spitting out sexualized deepfakes.
・| Image: Cath Virginia / The Verge | Photos from Getty Images Hugging Face is being used to make nonconsensual deepfakes, and the popular open-source AI model repository is doing very little to prevent it.
・That's according to a new report published by the European nonprofit AI Forensics, which found that seven out
cs.LG updates on arXiv.org
Impute On-Demand: Adaptive Correlated Time Series Imputation for Changing Environments
・arXiv:2607.23503v1 Announce Type: new Abstract: Internet of Things (IoT) applications generate vast amounts of Correlated Time Series (CTS) data that often contain missing values and require imputation.
・Existing methods emphasize accuracy but often lack adaptability to changing IoT environments: they are vulnerable to sensor failures, cannot selectively impute only incomplete sensors, and use static architectures tha
cs.LG updates on arXiv.org
In-Context Learning as Implicit Policy Gradient
・arXiv:2607.23153v1 Announce Type: new Abstract: Recent work has shown that large language models (LLMs) can iteratively improve their outputs by incorporating generated samples and their corresponding evaluation scores as in-context examples.
・Despite these empirical findings, the theoretical foundations underlying this phenomenon remain poorly understood.
・In this paper, we show that score-conditioned In-Context Learn
Hugging Face Papers
IndicTalk: A Large-Scale Persona-Based Multilingual Conversational Corpus for Indic Languages
IndicTalk: A Large-Scale Persona-Based Multilingual Conversational Corpus for Indic Languages
cs.LG updates on arXiv.org
IndicTalk: A Large-Scale Persona-Based Multilingual Conversational Corpus for Indic Languages
・arXiv:2607.23242v1 Announce Type: cross Abstract: Large Language Models (LLMs) have transformed conversational AI, yet high-quality multilingual code-mixed dialogue resources remain scarce, particularly for Indic languages where speakers naturally alternate between English and their native language in both native-script and Romanized forms.
・We present IndicTalk, one of the largest multilingual Indic code-mixed conver
cs.LG updates on arXiv.org
Invariant Discovery for Networked Systems
・arXiv:2607.22944v1 Announce Type: cross Abstract: Invariants, the relations expected to hold among measured signals of a network, underpin applications from verification to traffic generation, telemetry imputation, and input validation, yet writing them by hand demands rare expertise in both formal logic and networking.
・Automatic miners can help but fall short on two fronts: they still require the hardest input (the
WIRED
Is the Electric Trike the Next Big Thing in Shared Micromobility?
・Veo is launching accessibility-focused electric tricycles in Denver, with plans to put its e-trikes on the streets nationwide.
Hugging Face Papers
JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents
JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents
cs.LG updates on arXiv.org
Joint Flow Matching for Generator-Consistent Classification
・arXiv:2607.23946v1 Announce Type: new Abstract: We introduce Joint Flow Matching (JFM), a training framework for continuous normalising flows over multiple variables.
・Standard flow matching transports variables from noise to data simultaneously, offering no natural mechanism for forward and reverse conditional inference from a shared joint model.
・JFM resolves this by assigning opposite roles to each variable at the t
cs.LG updates on arXiv.org
K-Survival Means
・arXiv:2607.24405v1 Announce Type: new Abstract: In this work, we propose K-SurvMeans, a novel extension of K-Means for clustering survival data.
・The method explicitly uses the survival outcome in the clustering process to optimize cluster centers, thereby maximizing pairwise survival differences between clusters.
・The objective function encourages the clusters to be well-separated from the survival perspective.
cs.LG updates on arXiv.org
KAP: Bridging the Knowledge Selection-Runtime Consumption Gap in LLM Systems
・arXiv:2607.24260v1 Announce Type: new Abstract: Modern LLM systems increasingly rely on knowledge-selection processes that produce high-value structured priors, such as ranked evidence, graph topology, multimodal alignment, and confidence signals.
・Yet LLM serving remains fundamentally oblivious to this rich structure: once such signals are serialized into a prompt, the backend observes only a flat token sequence, for
#LLMタグ
Kimi K3で制作は「生成」から「反復」へ / Catch up on AI 2026.7.28
・Kimi K3は、総パラメータ数2.8兆のMoEモデル。処理時には1040億パラメータを有効化し、画像を扱うネイティブなマルチモーダル機能と、最大100万トークンのコンテキストを備えています。モデルウェイトと技術レポートも公開されました。
・Kimi K3が、モデルウェイトと技術レポートを公開。
・ネイティブな視覚理解と100万トークンのコンテキストを備え、スクリーンショットや制作物を見ながら、コードやツールを使った長時間の作業を進められる設計。
cs.LG updates on arXiv.org
Label-free Industrial Fault Detection via Adversarial Inverse Reinforcement Learning: A System for Run-to-Failure Prognostics
・arXiv:2607.22987v1 Announce Type: new Abstract: Machinery fault detection (MFD) remains heavily reliant on supervised learning, which struggles with the scarcity of fault labels in real-world settings.
・While reinforcement learning (RL) offers a framework to model the sequential nature of degradation, current ``RL-based'' MFD methods reduce the problem to a static contextual bandit (CB) formulation: by ignoring state
cs.LG updates on arXiv.org
Latent-LoRA: Compact Latent-Space Adapters with Gradient-Free Routing for Continual Learning
・arXiv:2607.23837v1 Announce Type: new Abstract: Large language models generalize well to individual tasks but lack an inherent mechanism for learning them sequentially, leading to catastrophic forgetting.
・To mitigate this, LoRA-based continual learning methods allocate a separate low-rank adapter per task, yet existing approaches either require task identity at inference or sum all adapters indiscriminately, letting
cs.LG updates on arXiv.org
LC-SEPLM: long-range contact-supervised adaptation for sequence-only protein representation learning
・arXiv:2607.22777v1 Announce Type: new Abstract: Protein language models learn transferable sequence representations.
・However, because they primarily model contextual dependencies along amino-acid sequences, their training objectives do not explicitly constrain the model to learn three-dimensional residue contacts formed after folding .
・Here, we introduce LC-SEPLM (Long-range Contact-supervised ESM Protein Language Mo
cs.LG updates on arXiv.org
Learned Interventions in Lean 4 grind
・arXiv:2607.22972v1 Announce Type: new Abstract: Lean~4's \grind{} tactic combines congruence closure, \ematch{}ing, and case-splitting into a single automated solver, and like any such solver, it relies on hand-tuned heuristics to decide what to instantiate and where to case-split.
・These heuristics are tempting targets for learning, but there is a catch: because \grind{}'s search is non-monotone, a learned heuristic
cs.LG updates on arXiv.org
Learning from the Descent Direction: Adaptive Gradient Descent under One-Sided H\"older Regularity
・arXiv:2607.22906v1 Announce Type: new Abstract: We study adaptive gradient descent for continuously differentiable, possibly nonconvex objectives under one-sided H\"older regularity.
・Unlike classical H\"older- or Lipschitz-gradient assumptions, which control the full gradient variation, our condition bounds only the directional term appearing in the descent inequality.
・This can allow less conservative step sizes when
cs.LG updates on arXiv.org
Learning Reusable Hybrid Motion Priors for Humanoid Locomotion from Motion Imitation
・arXiv:2607.24083v1 Announce Type: new Abstract: Reinforcement learning can produce robust humanoid controllers, but each new task is typically trained as a separate policy with its own reward design and training process.
・Motion imitation provides an alternative source of motor competence by training policies to track retargeted human motions, yet the resulting controllers remain reference trackers and are not directl
cs.LG updates on arXiv.org
Learning Sampling Parameters for Diffusion Models
・arXiv:2607.23488v1 Announce Type: new Abstract: Text-to-image diffusion models expose many inference-time sampling parameters, including prompts, negative prompts, classifier-free guidance scales, and noise schedules.
・These parameters are typically manually chosen once and then held fixed across prompts and denoising timesteps, even though different prompts and stages of generation can benefit from different paramete
cs.LG updates on arXiv.org
Learning to Access Computation: Accessibility Plasticity as a Principle of Adaptive Intelligence
・arXiv:2607.22748v1 Announce Type: new Abstract: Modern neural networks primarily adapt through parameter modification within predefined computational structures.
・While recent methods introduce modularity, conditional computation, and parameter-efficient adaptation, they generally do not distinguish computational capability from computational accessibility as separate adaptive variables.
・This work introduces Accessibi
cs.LG updates on arXiv.org
Learning to Optimize at Scale: A Benders Decomposition-TransfORmers Framework for Stochastic Combinatorial Optimization
・arXiv:2607.22550v1 Announce Type: cross Abstract: We propose a learning-augmented Benders decomposition framework to solve large-scale two-stage stochastic mixed-integer programs.
・We focus on the two-stage stochastic capacitated lot-sizing problem (TSSCLSP) under demand uncertainty.
・Our method accelerates the convergence of the decomposition by using a pre-trained TransfORmer model to rapidly generate high-quality ap
cs.LG updates on arXiv.org
Learning to Optimize: Joint Routing and Flow Allocation on Sparse Non-Euclidean Networks
・arXiv:2607.23467v1 Announce Type: new Abstract: We study an integrated pickup-and-delivery problem on sparse, non-Euclidean networks that jointly optimizes cyclic routing, cargo flow allocation, and cross-cycle service.
・The tight coupling of these operational constraints creates a complex discrete-continuous decision space with highly restricted feasible regions.
・To overcome these computational challenges, we propose
Hugging Face Papers
Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models
Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models
Hugging Face - Blog
LFM2.5-Encoders for Fast Long-Context Inference on CPU
LFM2.5-Encoders for Fast Long-Context Inference on CPU
cs.LG updates on arXiv.org
LithoFormer: A Robust Framework for Stratigraphic Inference via Transformers
・arXiv:2607.22804v1 Announce Type: new Abstract: Accurate geological characterization of subsurface reservoirs from well log data is essential to support projects such as carbon capture and storage (CCS), geothermal development, and extraction of natural resources.
・Existing automated techniques for geological characterization primarily use sliding-window classification, which limits their ability to understand broader
cs.LG updates on arXiv.org
Local Regularization Does Not Characterize Multiclass PAC Learnability
・arXiv:2607.23449v1 Announce Type: new Abstract: Local regularization assigns each hypothesis a test-point-dependent score and predicts with a minimum-score hypothesis consistent with the sample.
・Asilis et al.
・asked whether this principle characterizes multiclass PAC learnability.
cs.LG updates on arXiv.org
LOCKS: Page-Local Compact Key Summaries for Efficient Long-Context Decoding
・arXiv:2607.24555v1 Announce Type: new Abstract: Serving large language models at long context is bottlenecked by the key-value (KV) cache, which is read in full at every decode step.
・Attention keys are locally low-rank though globally high-rank: shared low-rank bases discard page-specific directions that a page's own compact basis retains.
・LOCKS gives every page its own spectral summary (resident, about a tenth the c
The Verge
Logitech’s handheld plans are on ice — don’t expect a G Cloud 2 soon
・Logitech's new gaming boss, Robin Piispanen, tells me he likes the idea of gaming handhelds.
・"It's such a charming value proposition," he says, as we sip iced vanilla lattes at my local cafe.
・But he's not building one right now.
cs.LG updates on arXiv.org
Low-Rank Dependence Decomposition via Accelerated Symmetric Non-negative Matrix Factorization
・arXiv:2607.24518v1 Announce Type: new Abstract: Symmetric non-negative matrix factorization (SymNMF) recovers latent group structure from a dependence matrix, but its dense, quadratic-memory objective has confined prior work to moderate sizes.
・We present a large-scale GPU study of seven algorithm families (over 30 configurations) on absolute Pearson correlation and tail pairwise dependence matrices from Extreme Value
cs.LG updates on arXiv.org
MAPLE: Efficient and Diverse Multi-Alpha Generation for Portfolio Construction
・arXiv:2607.24131v1 Announce Type: new Abstract: Classical alpha mining achieves strong risk-adjusted returns by combining many low-correlated predictive signals, yet deep learning stock-ranking methods typically produce a single alpha per stock, rely on increasingly complex architectures with diminishing gains, and obtain diversity only through separate models or implicit routing, without explicitly controlling inter
cs.LG updates on arXiv.org
Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data
・arXiv:2607.22615v1 Announce Type: cross Abstract: Clinical neuroprosthetics face a data bottleneck: labeled perception trials are scarce while hours of spontaneous neural activity are largely underutilized.
・Here, we test whether self-supervised learning can use these unlabeled datasets to improve perception decoding.
・We pretrained a masked autoencoder on 14.6 hours of spontaneous multiunit activity from an intracorti
cs.LG updates on arXiv.org
MEGA-CL: A Molecular Foundation Model for Generalizable ADMET Prediction through Graph External Attention and Contrastive Learning
・arXiv:2607.24314v1 Announce Type: new Abstract: Predicting the absorption, distribution, metabolism, excretion and toxicity (ADMET) properties of small molecules remains a major challenge in drug discovery.
・Here, we present MEGA-CL, a foundation graph neural network framework for universal molecular ADMET prediction.
・MEGA-CL integrates self-supervised contrastive learning with a multi-head external attention mechanis
cs.LG updates on arXiv.org
MEMENTO: Memory-Guided Memetic Code-as-Policy Evolution
・arXiv:2607.22832v1 Announce Type: new Abstract: Long-horizon embodied tasks require policies that execute many dependent actions before task success can be observed.
・Representing policies as executable control pro- grams (code-as-policy) enables their decision logic to be inspected and revised after rollout evaluation.
・Revised programs can then be executed and compared by rollout performance, framing policy improveme
cs.LG updates on arXiv.org
Meshless Domain Randomization via Explicit Parameter Perturbation of 3D Gaussian Splatting
・arXiv:2607.22890v1 Announce Type: cross Abstract: Domain Randomization (DR) is a standard technique for closing the Sim-to-Real gap, yet traditional DR pipelines rely on classical computer graphics rendering driven by polygon meshes.
・For complex organic subjects, such as insect specimens, extracting and rendering textured meshes is challenging.
・To address this issue, we propose a meshless DR framework that operates o
cs.LG updates on arXiv.org
Metamorphic Testing for Clinical ML Models: A Framework Proposal and Pilot Study
・arXiv:2607.22984v1 Announce Type: cross Abstract: Machine learning models for clinical prediction tasks, such as in-hospital mortality and sepsis onset, routinely achieve high AUROC scores.
・However, AUROC measures ranking performance rather than clinical sensibility.
・A model may rank patients correctly overall while predicting a lower mortality risk when a patient's SOFA score worsens, contradicting established medic
MarkTechPost
Microsoft AI Releases MAI-Cyber-1-Flash: A 5B-Active-Parameter Cyber Model That Pushes MDASH to 95.95% on CyberGym
・Microsoft AI has released MAI-Cyber-1-Flash, its first model built specifically for cyber defense.
・It is a 137B total, 5B active sparse MoE fine-tune of MAI-Code-1-Flash with a 256k context window.
・The model does not ship as a standalone endpoint — it runs inside MDASH, Microsoft's multi-model agentic scanning harness, where it handles up to 90% of tasks and pushes the system to 95.95% on CyberGym.
cs.LG updates on arXiv.org
MIME: Multimodal Interactive Motion Encoder
・arXiv:2607.22702v1 Announce Type: cross Abstract: Text-motion representation learning has advanced rapidly, with growing interest in multi person interactions for animation, AR/VR, and embodied AI.
・These settings require representations that align language with both individual actor dynamics and the relationships between actors.
・We introduce the Multimodal Interactive Motion Encoder (MIME), which, to our knowledge, r
cs.LG updates on arXiv.org
Mini-batch Noise Lowers Sharpness via Dominant-Subspace Fluctuations
・arXiv:2607.23012v1 Announce Type: new Abstract: During SGD training, the gradients often align strongly with the dominant subspace spanned by the top-$k$ eigenvectors of the Hessian of the loss.
・While this seems to naturally imply that loss reduction mainly occurs within this space, prior work has shown that updates within this dominant subspace make no meaningful progress in reducing the loss.
・In this work, we argue
cs.LG updates on arXiv.org
MioFFAn: an Annotation Software for Formula Formalization with LLM Automation Capabilities
・arXiv:2607.22552v1 Announce Type: cross Abstract: The automatic translation of mathematical expressions in scientific literature into executable symbolic code (a process we refer to as Formula Formalization) is hindered by a severe scarcity of high-quality, ground-truth datasets specialized for technical scientific domains.
・In this paper, we present MioFFAn, an open-source, document-centric, and customizable framewor
cs.LG updates on arXiv.org
MixQuant: Adaptive Mixed-Precision Quantization for Large Language Models
・arXiv:2607.23047v1 Announce Type: new Abstract: Mixed-precision quantization improves the accuracy of post-training quantization by allocating higher bitwidths to sensitive layers, but existing methods solve the allocation for a single fixed memory budget.
・In practice the budget varies across deployments and is unknown at calibration time.
・Adaptive quantization addresses this with one offline calibration that serves
cs.LG updates on arXiv.org
ML-based Predictive Models for Power Consumption in Virtualised O-RANs
・arXiv:2607.24256v1 Announce Type: new Abstract: As communication networks adopt virtualized and disaggregated architectures, achieving energy efficiency has become increasingly important for both economic and environmental reasons.
・Traditional methods for power modeling are inadequate in these dynamic software-defined environments due to their inability to model complex and nonlinear factors affecting energy use.
cs.LG updates on arXiv.org
MobiWave: Dispatch-Oriented Graph Wavelets and Drift-Guided Selective Optimization for Autonomous Fleet Rebalancing
・arXiv:2607.24365v1 Announce Type: new Abstract: Autonomous fleets enable mobility platforms to coordinate idle vehicles directly, making fleet-wide rebalancing possible.
・However, two obstacles limit reliable deployment: overlapping regional and local traffic patterns can hide roads that remain useful for dispatch, and mobility drift can make a trained policy unreliable.
・Existing spatial aggregation mixes these patter
cs.LG updates on arXiv.org
Modeling Memory-Dependent Reliability of LLMs: A Hidden Markov Model
・arXiv:2607.22951v1 Announce Type: cross Abstract: Reliability assessment of large language models (LLMs) seeks to estimate the probability that a model produces correct responses under a specified operational profile.
・Conventional benchmark-based evaluation, often summarized by aggregate accuracy, provides a point estimate of performance but does not characterize the uncertainty associated with reliability claims.
cs.LG updates on arXiv.org
Monitoring Post-Disaster Urban Recovery Using High-Resolution SAR Time Series and Unsupervised Learning: Evidence from the 2023 T\"urkiye-Syria Earthquake
・arXiv:2607.24180v1 Announce Type: new Abstract: Monitoring post-disaster recovery is essential for understanding how urban systems rebuild and progressively return to functionality.
・However, tracking reconstruction remains difficult because reliable ground-truth information is often scarce and recovery processes evolve over time.
・This paper proposes an unsupervised framework for recovery monitoring based on multi-tem
cs.LG updates on arXiv.org
MS-GPT: Rethinking MS/MS De Novo Structure Elucidation as Spectrum-Induced Posterior Querying of a Molecule-Language Model
・arXiv:2607.23607v1 Announce Type: new Abstract: Molecular structure elucidation from tandem mass spectra (MS/MS) is a central inverse problem in analytical chemistry.
・Most existing approaches to MS/MS identification remain tied to reference libraries or predefined candidate sets, whereas de novo methods aim to generate structures directly from spectra.
・A common de novo route predicts a molecular fingerprint from the
cs.LG updates on arXiv.org
Multi-Agent Privacy Game in Federated Learning: A Unified Mean-Field View
・arXiv:2607.23029v1 Announce Type: new Abstract: Federated learning enables collaborative model training across distributed clients without centralising their data, yet privacy remains a persistent concern because the shared model updates can leak information about local datasets.
・Existing privacy-preserving methods either inject calibrated noise into client updates, limiting their composition guarantees, or formulate
cs.LG updates on arXiv.org
Multimodal Domain Generalization for Depression Detection: An Attention-Based BiLSTM Network with Domain-Adversarial Training
・arXiv:2607.22794v1 Announce Type: new Abstract: Automatic depression detection with deep learning has shown promise but often suffers from limited generalization due to domain shift arising from inter-speaker variability.
・To address this critical issue, we present the first patient-independent multimodal depression detection framework that incorporates domain generalization (DG), jointly leveraging both acoustic and
cs.LG updates on arXiv.org
Multimodal Surface EMG Hand Gesture Recognition Using Query-Based Transformers for Prosthetic Control
・arXiv:2607.22779v1 Announce Type: new Abstract: Hand gesture recognition via surface electromyography (sEMG) is fundamental to prosthetic control.
・In this field, deep learning approaches have become the gold standard.
・However, current architectures struggle to scale; model performance typically decreases as the number of hand movements increases.
cs.LG updates on arXiv.org
MXAttention: Data-Free Optimal Scaling and Pre-Normalization Quantization for MXFP4 Attention
・arXiv:2607.24377v1 Announce Type: new Abstract: The quadratic cost of attention is a major bottleneck in diffusion-based video generation models.
・MXFP4 attention provides a promising path toward efficient inference, but direct MXFP4 quantization often degrades generation quality due to two numerical issues: the clipping-underflow trade-off from power-of-two scaling and the row-wise normalization error introduced in t
cs.LG updates on arXiv.org
Neonatal Hypoxic-ischaemic Encephalopathy Classification from the EEG and HRV Signals Using a Conformer based Masked Autoencoder
・arXiv:2607.23554v1 Announce Type: new Abstract: In this paper, we propose the MAEConformer, a novel self-supervised learning framework that combines the Conformer architecture with the Masked Autoencoder (MAE) paradigm for large-scale representation learning from unlabelled electroencephalography (EEG) and heart rate variability (HRV) signals.
・By integrating convolutional operations with Transformer-based self-attent
cs.LG updates on arXiv.org
Nesterov acceleration in optimizing over probability measures
・arXiv:2607.23008v1 Announce Type: cross Abstract: Optimization over probability measures has become an increasingly important paradigm in modern machine learning, scientific computing, and uncertainty quantification.
・Motivated by Nesterov's accelerated gradient method in Euclidean space, we develop Heavy-ball and Nesterov acceleration methods over the probability measure space $\mathcal{P}_2$ and establish non-asympt
cs.LG updates on arXiv.org
Neural Network-Driven Volatility Drag Mitigation under Aggressive Leverage
・arXiv:2607.23068v1 Announce Type: cross Abstract: This paper introduces a compact reformulation of a modular end-to-end neural network for global minimum-variance portfolio optimization that decouples model complexity from both look-back window length and universe size.
・A five-parameter hyperbolic weighted moving average combined with a saturating exponential replaces the original 2,400-parameter lag-transformation l
cs.LG updates on arXiv.org
Neural operator discovery from heterogeneous trajectories
・arXiv:2607.23337v1 Announce Type: new Abstract: Neural operators provide data-driven mappings for modeling dynamical systems.
・Extending them to families of systems typically requires explicit conditioning variables such as physical parameters, geometries, or boundary conditions.
・In many real-world settings, these quantities are unobserved.
ITmedia NEWS 最新記事一覧
NHKのネットサービスが受信契約なしで閲覧可能に 熊本地震受け緊急対応
・7月28日午後4時27分ごろに発生した「令和8年熊本地震」を受け、NHKのインターネットサービス「NHK ONE」が、誰でも閲覧できるようになっている。通常は受信契約者のみアクセス可能だが、緊急対応として防災情報などを公開している。
cs.LG updates on arXiv.org
Not All LLM Reasoning is Visible in the Chain-of-Thought
・arXiv:2607.22925v1 Announce Type: cross Abstract: A key question for AI safety is whether a language model expresses all of its reasoning in its output tokens.
・We demonstrate a concrete failure mode where frontier models exhibit invisible reasoning by leveraging semantically irrelevant filler tokens to improve performance on synthetic reasoning tasks.
・We evaluate 13 frontier language models across three tasks and fin
Hugging Face Papers
OmniVAE: An Audio-Video VAE with Cross-Modal Alignment for Joint Generation
OmniVAE: An Audio-Video VAE with Cross-Modal Alignment for Joint Generation
cs.LG updates on arXiv.org
On the Impossibility of Unbiased and Length-Invariant Policy Optimization with Outcome Rewards
・arXiv:2607.23364v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) is the dominant reinforcement learning algorithm for training reasoning capabilities in large language models, notably adopted by DeepSeek-R1.
・The recent improvement Dr.
・GRPO (COLM 2025) identifies the response-level length bias caused by per-trajectory length normalization in GRPO and proposes removing this normalization, claim
cs.LG updates on arXiv.org
On the Order-Conditional Optimality of Gaffke's Bound
・arXiv:2607.22971v1 Announce Type: cross Abstract: Let $X = (X_1, \ldots, X_n)$ be a random vector from any Borel probability law on $\mathbb{R}_+^n$.
・We revisit the problem of deriving a lower confidence bound (LCB) on a scalar parameter of that law.
・We recast classical work, beginning with Buehler, in purely probabilistic terms to form a more accessible and extensible framework.
cs.LG updates on arXiv.org
On the post-hoc Evaluation of PDE Discovery: A Multifaceted Challenge of Scientific Advancement
・arXiv:2607.23753v1 Announce Type: new Abstract: Partial differential equation (PDE) discovery aims to identify from data the governing law of a physical system.
・Constituting a cornerstone of scientific advancement, it has become during the past decade a major line of research in the rapidly evolving field of Physics-informed Machine Learning (PiML).
・Among the remaining open problems to address in this domain, the pos
cs.LG updates on arXiv.org
Online Policy Evaluation for MDPs with Dynamic UBSR Measures
・arXiv:2607.23030v1 Announce Type: new Abstract: Developing efficient function-approximation methods for policy evaluation is a fundamental challenge in risk-aware reinforcement learning.
・Existing approaches either focus on restrictive classes of risk measures or rely on access to a simulator, limiting their applicability in fully online settings.
・In this work, we propose computationally efficient online learning algo
cs.LG updates on arXiv.org
Operator Neural Jump ODEs: $L^2$-optimal prediction in function spaces
・arXiv:2607.23110v1 Announce Type: cross Abstract: In this paper, we study the extension of Neural Jump ODEs to infinite-dimensional function spaces.
・In particular, the underlying process $X$ now takes values in $L^2(\Xi, \mathbb{R}^{d_X})$ instead of $\mathbb{R}^{d_X}$ and the Operator NJ-ODE approximates the optimal predictor of this process by producing a representative of the conditional expectation.
・The NJ-ODE mo
cs.LG updates on arXiv.org
Optimal Reward Shaping: Autonomous Car Parking Case Study
・arXiv:2607.23617v1 Announce Type: new Abstract: Designing effective reward functions for model-free reinforcement learning under non-holonomic constraints remains a persistent challenge, often resulting in severe local minima such as policy paralysis or over-conservative hazard avoidance.
・In this work, we present a parameterized reward shaping framework featuring coverage-gated alignment feedback, drive-direction swi
cs.LG updates on arXiv.org
Optimizing Transformer Neural Network for Real-Time Outlier Detection on FPGAs
・arXiv:2607.22786v1 Announce Type: new Abstract: In this work, we explore how the inference time of a Transformer Neural Network can be efficiently optimized with applications to real-time anomaly detection in financial time series.
・The financial time series are price series such as asset prices.
・Unfortunately, the data is often with errors or outliers that make the downstream data processing tasks useless, unstable o
cs.LG updates on arXiv.org
OrchNAS: Orchestrated Neural Architecture Search Service for Personalised Federated Edge Intelligence
・arXiv:2607.22805v1 Announce Type: new Abstract: We propose OrchNAS, an energy-aware, personalised, federated edge intelligence framework that leverages a Neural Architecture Search Service to automatically design service-adaptive models for heterogeneous edge environments.
・The framework orchestrates the architecture search process on a server-side NAS service, enabling edge services to derive personalised architectur
cs.LG updates on arXiv.org
Outcome-Confounded Local Supervision in On-Policy Distillation
・arXiv:2607.23731v1 Announce Type: new Abstract: On-policy distillation (OPD) trains a student on its own trajectories while a teacher supplies dense token-level likelihoods at student-visited prefixes.
・These likelihoods are often read locally: agreement appears safe to imitate, whereas disagreement appears to identify an error.
・We show that both readings are confounded by the outcome of the completed trajectory.
Hugging Face Papers
Oxygen-TryOn: Fashion-Native Foundation Model for Any-item Virtual Try-On
Oxygen-TryOn: Fashion-Native Foundation Model for Any-item Virtual Try-On
cs.LG updates on arXiv.org
ParasGB: A Graph Benchmark Suite for Parasitic Estimation on AMS Circuits
・arXiv:2607.23225v1 Announce Type: new Abstract: As chip manufacturing processes advance to deep submicron nodes, parasitic interconnect effects increasingly dominate the performance of analog and mixed-signal (AMS) circuits and often lead to costly layout iterations.
・This makes early-stage estimation of parasitic capacitance and resistance important for parasitic-aware design exploration before full physical implemen
The Verge
Perplexity’s Personal Computer turns Windows PCs into AI agents
・Perplexity has expanded its agentic Personal Computer tool to Windows, allowing computers running the world's most popular OS to be used as a locally run AI system.
・Like the Mac version that Perplexity launched in April, Personal Computer for Windows operates like a "general-purpose digital worker" that can access local files and apps to perform actions on your behalf, such as creating documents and updating spreadsh
cs.LG updates on arXiv.org
Perturbative-NeuSA: A Structured Spectral Framework for Time-Dependent PDEs
・arXiv:2607.24345v1 Announce Type: new Abstract: Neural spectral PDE solvers often learn an entire unresolved vector field even when an inexpensive approximate model can already capture most of the trajectory.
・Here we introduce Perturbative-NeuSA, a residual formulation that decomposes the target solution into a low-fidelity background and a high-resolution perturbation, so that only the unresolved dynamics is learned
cs.LG updates on arXiv.org
PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling
・arXiv:2607.23447v1 Announce Type: new Abstract: Predicting cellular responses to unseen chemical perturbations is challenging due to unknown targets and mechanisms, high-dimensional expression responses, and limited experimental coverage of the large small-molecule design space.
・We propose PerturbPFN, a PFN-style amortized model for unknown-target perturbation prediction under a hierarchical synthetic structural prio
cs.LG updates on arXiv.org
Physically Verifiable Evidence and LLM-Based Reporting for Bearing Fault Diagnosis
・arXiv:2607.22797v1 Announce Type: new Abstract: Trustworthy deployment of AI-based diagnosis in safety-critical mechanical systems hinges on validation: whether a prediction can be checked against physical reality before it is acted upon.
・Current intelligent fault diagnosers fail this standard in two ways.
・Their standard output, a class label with a softmax confidence score, is an internal statistic of the classifier
cs.LG updates on arXiv.org
Physics Transformer: Tailoring Transformer for General PDE Prediction
・arXiv:2607.24513v1 Announce Type: new Abstract: Transformer architectures have attracted increasing attention for solving partial differential equations (PDEs), owing to their flexibility in handling irregular discretizations and their ability to capture long-range physical dependencies.
・However, unlike discrete language tokens or fixed-resolution image patches, observed physical fields are finite samples of underlyi
cs.LG updates on arXiv.org
Physics-Guided Generative AI for Property-Targeted 3D Porous Media Design
・arXiv:2607.24274v1 Announce Type: new Abstract: Inverse design of three-dimensional porous media is central to applications in filtration, catalysis, energy storage, fuel cells, thermal management, and biomedical scaffolds, but remains challenging because many distinct pore geometries can share similar porosity or permeability while small structural changes can strongly affect transport behaviour.
・This paper proposes
cs.LG updates on arXiv.org
Physics-Informed Neural Networks for Discovering Periodic Orbits in the Gravitational Three-Body Problem
・arXiv:2607.23501v1 Announce Type: new Abstract: Locating periodic solutions of chaotic dynamical systems normally requires an initial guess close enough to the target orbit for numerical continuation or gradient-based search to converge.
・We show that Physics-Informed Neural Networks (PINNs) trained on sparse, noisy observations \emph{without} initial conditions recover periodic orbits of the gravitational three-body
cs.LG updates on arXiv.org
Physics-Informed Neural Networks for Predicting Nitrous Oxide Flux
・arXiv:2607.23880v1 Announce Type: new Abstract: Nitrous oxide (N$_2$O) is the dominant ozone-depleting substance emitted in the 21st century, and the third largest contributor to anthropogenic greenhouse gases due to its high potency and long atmospheric lifetime, with more than 70% of N$_2$O emissions occurring as a result of agricultural processes.
・Current approaches to predicting N$_2$O flux emissions include proc
NVIDIA Blog
Powerful Compute So Compact, It’s Clutch — Build AI Anywhere With NVIDIA Jetson
・As a discerning AI investor who values style and substance, Sarah Guo knows this season’s standout accessory isn’t the latest designer purse — but what’s inside it.
・In a recent video, Guo, founder of AI-native venture capital firm Conviction and co-host of the AI podcast No Priors, highlighted how the NVIDIA Jetson platform for edge […]
cs.LG updates on arXiv.org
Practical advantage beyond the quadratic speedup limit with fully-quantum walks
・arXiv:2607.22818v1 Announce Type: cross Abstract: We introduce a new class of fully-quantum Metropolis walks in which both the proposal and acceptance steps are intrinsically quantum.
・Unlike standard quantum walks obtained by quantizing classically efficient Markov chains, our algorithm employs Hamiltonian simulation as a quantum-native proposal mechanism, enlarging the class of quantum walks beyond classical counter
cs.LG updates on arXiv.org
Predicting the Outcome of rTMS Depression Therapy using EEG Signals and CNN
・arXiv:2607.22776v1 Announce Type: new Abstract: Repetitive transcranial magnetic stimulation (rTMS) is a non invasive therapy for Major Depressive Disorder (MDD).
・In this study, we generate images using two time frequency methods to represent EEG signals: Fourier-Bessel Series Expansion with Euclidean Distance (FBSE-ED) and Discrete Wavelet Transform (DWT).
・We propose an efficient deep learning classifier to predict
Hugging Face Papers
Progress Reward Modeling for Robotic Learning: A Comprehensive Survey
Progress Reward Modeling for Robotic Learning: A Comprehensive Survey
cs.LG updates on arXiv.org
Progress-conditioned Group Policy Optimization for Long-Horizon Agentic Tasks
・arXiv:2607.22724v1 Announce Type: new Abstract: Group-based policy optimization has been increasingly used to train large language model (LLM) agents from sparse outcome rewards by comparing trajectories or steps within a group.
・However, on difficult long-horizon tasks, this comparison can suffer from a sampling imbalance: repeated or low-effect actions dominate the high-probability region of the policy while useful
cs.LG updates on arXiv.org
PYPM-GGD: Pitman-Yor Process Mixture with Generalized Gaussian Density using ADAM
・arXiv:2607.24583v1 Announce Type: new Abstract: Large scale Bayesian nonparametrics (BNP) learner such as Stochastic Variational Inference (SVI) can handle datasets with large class number and large training size at fractional cost.
・Like its predecessor, SVI rely on the assumption of conjugate variational posterior to approximate the true posterior.
・A more challenging problem is to consider large scale learning on no
cs.LG updates on arXiv.org
QFedPolyp: A Communication- and Inference-Efficient Federated Learning Framework for Polyp Segmentation
・arXiv:2607.22743v1 Announce Type: new Abstract: Background and Objective: Automatic polyp segmentation supports computer-aided diagnosis and early colorectal cancer detec- tion.
・Centralized deep learning requires hospitals to share sensitive medical data, while federated learning preserves privacy but introduces high communication costs through repeated transmission of full-precision model parameters.
・We propose QFed
cs.LG updates on arXiv.org
Queryable Self-Organizing Maps: A Database Abstraction for Topology-Driven Data Exploration
・arXiv:2607.22843v1 Announce Type: cross Abstract: Self-Organizing Maps (SOMs) have long been used as exploratory tools for high-dimensional data: they organize objects into a two-dimensional topology that reveals clusters, gradients, sparse regions, dense regions, and boundaries.
・Yet, in modern data systems, SOMs are typically trained and visualized outside the DBMS, disconnected from the relational data they summari
cs.LG updates on arXiv.org
Quotient Tree Arithmetic: Deferred-Division Computation with Bounded Symbolic Depth and Cross-Subtree Cancellation
・arXiv:2607.22612v1 Announce Type: cross Abstract: We introduce Quotient Tree Arithmetic (QTA), a computational substrate in which values are represented as deferred quotient pairs (N, D) whose ratio is evaluated lazily at a designated materialization boundary.
・The framework applies to any domain: IEEE 754 doubles used as exact integer containers give exact rational arithmetic within the 2^53 exactness window; arbitra
cs.LG updates on arXiv.org
Random Forest-Based Prediction of Bone Volume Fraction and Fracture Position from S-Parameters
・arXiv:2607.23563v1 Announce Type: new Abstract: In this paper, we propose a method for predicting bone volume fraction (BVF) and fracture position by constructing a random forest model based on multichannel S-parameters.
・A nine-antenna microwave scanning system is designed and fabricated to acquire the multichannel S-parameter data.
・Bone-mimicking phantoms are developed, and corresponding experiments are conducted to
WIRED
Ray-Ban Promo Codes: Save 50% in July 2026
・Upgrade your frames with major savings on classic shapes, custom designs, and prescription lenses using these verified Ray-Ban discounts.
cs.LG updates on arXiv.org
Real2Sim2Real for Vision-Language-Action Manipulation: An AMD ROCm-Based Pipeline
・arXiv:2607.22997v1 Announce Type: cross Abstract: Physical AI -- the integration of large vision-language-action (VLA) models with embodied agents that act in the real world -- has emerged as the next major frontier for AI, echoed by industry leaders such as Jensen Huang (``the next big thing is Physical AI, AI with a body,'' GTC Paris, June 2025) and Dr.
・Lisa Su (`we're entering the world of Physical AI ...
Hugging Face Papers
Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models
Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models
AI News & Artificial Intelligence | TechCrunch
Recursive Superintelligence signs $410M compute deal with Amazon
・Recursive’s emphasis on self-improving AI systems means much of the budget that would traditionally go toward headcount and operations is put straight into compute, as the company seeks to automate its own product development process.
cs.LG updates on arXiv.org
Recycling computational processes of dynamic programming for combinatorial optimization problems: a reservoir computing approach
・arXiv:2607.23009v1 Announce Type: new Abstract: Reusing previously computed results is a long-standing principle for reducing computational cost, but such reuse has largely been confined to a single problem's computation.
・Sharing computational processes across multiple simultaneously solved problems remains possible in principle, yet designing algorithms that exploit nontrivial cross-task relationships is difficult t
cs.LG updates on arXiv.org
Reinforcement Learning for Heterogeneous Sensor Selection in Maritime Surveillance
・arXiv:2607.22667v1 Announce Type: 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-gain evalua
cs.LG updates on arXiv.org
Restoration Flow Matching-Based Channel Refinement and Equalization Correction for MIMO Semantic Communications
・arXiv:2607.23615v1 Announce Type: new Abstract: In multiple-input multiple-output (MIMO) semantic communication, imperfect channel state information (CSI) and equalization mismatch can seriously degrade semantic reconstruction quality.
・To address this issue, we propose a unified restoration flow matching (RFM)-based framework for channel refinement and equalization correction.
・Specifically, the channel RFM (CRFM) mod
Hugging Face Papers
Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
#LLMタグ
RIR・物理法則・拡散モデル・音響マルチポールから読む近年の到達点
・RIR・物理法則・拡散モデル・音響マルチポール オンライン会議で、相手の声が聞き取りにくい。
・ショールームにスピーカーを設置したものの、場所によって音の印象が大きく変わってしまう。
・工場では機械音が反響し、異常音検知の精度が安定しない。
cs.LG updates on arXiv.org
Robust Conformalized Selection with Noisy Responses
・arXiv:2607.22985v1 Announce Type: cross Abstract: Conformalized selection has been widely applied to select high-quality candidates from large datasets with rigorous uncertainty quantification, such as reliable labeling, drug discovery, and the alignment of large language models.
・Nevertheless, existing methods assume clean responses on calibration data, an assumption that rarely holds in practice.
・In this paper, we f
cs.LG updates on arXiv.org
Same Predictions, Different Reasons: The Effect of Quantization on Model Explanations
・arXiv:2607.22872v1 Announce Type: new Abstract: Post-training quantization (PTQ) has become a practical solution for deploying deep learning models on resource-constrained edge devices by compressing high-precision floating-point weights into low-precision representations without requiring retraining.
・Past research has demonstrated that quantization largely preserves classification accuracy; however, whether it also
cs.LG updates on arXiv.org
Same Question, Different Answers: Evaluating LLM Reliability Beyond Accuracy
・arXiv:2607.22554v1 Announce Type: cross Abstract: Large language models (LLMs) often achieve strong accuracy on benchmarks, yet it remains unclear how reliably they apply this knowledge when the same question is phrased in different but equivalent ways.
・In this work, we study how model answers change under meaning-preserving paraphrases across factual question answering and mathematical reasoning tasks.
・Across four b
cs.LG updates on arXiv.org
Scale Weight Decay and Train Better
・arXiv:2607.23777v1 Announce Type: new Abstract: The discovery of scaling laws has motivated training neural networks on ever increasing quantities of data.
・This is typically done with a constant decoupled weight decay which causes the network weights to shrink steadily over the course of training.
・Taking inspiration from the Robbins--Monro conditions, we propose to scale weight decay by the fraction of the peak learn
OpenAI News
Scientific computing in the age of agentic AI
・A new field report shows how scientists use AI coding agents to modernize scientific computing, accelerating software development and discovery in genomics and beyond.
cs.LG updates on arXiv.org
SCTA: An Agentic Framework for Stable and Interpretable Target Gene Discovery from Single-Cell RNA Sequencing
・arXiv:2607.23821v1 Announce Type: new Abstract: Identifying therapeutic target genes from single-cell RNA sequencing (scRNA-seq) data remains a fundamental challenge in translational biology.
・Unlike bulk assays, scRNA-seq captures heterogeneous cellular states and rare subpopulations, but this same heterogeneity makes target discovery highly sensitive to analytical choices throughout the pipeline, including preproces
cs.LG updates on arXiv.org
Self-Boosting Vision-Language Models with Noisy Student On-Policy Self-Distillation
・arXiv:2607.23125v1 Announce Type: new Abstract: Post-training enables vision-language models (VLMs) to understand human instructions and perform various downstream tasks.
・Current post-training methods usually rely on human-annotated data, distillation from external models, reinforcement learning with human feedback, or verifiable answers.
・This limits their ability to improve without external supervision.
cs.LG updates on arXiv.org
Semalith v1.4: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44x Fewer Parameters than Llama-Guard-3-8B
・arXiv:2607.22545v1 Announce Type: new Abstract: Deploying large language models in financial-services and agentic settings requires safety classifiers that simultaneously handle prompt injection, regulatory compliance, and general harm, a combination no existing open guardrail addresses in a single inference pass.
・Semalith v1.4 is a 184M-parameter DeBERTa-v3-base classifier performing simultaneous three-axis safety c
cs.LG updates on arXiv.org
SeT-Diff: Towards Semantic Foundation Models for HPC Telemetry and Time-Series
・arXiv:2607.22548v1 Announce Type: cross Abstract: Data centers and their compute nodes require accurate and flexible digital twins capable of modeling the complex interplay of workloads, environmental parameters, and physical metrics.
・Current machine learning approaches for HPC and its telemetry typically rely on a static subset of anonymous, fixed-position sensor variables tailored to single tasks.
・Consequently, the
Zennの「大規模言語モデル」のフィード
Sheaf Laplacianでデータの矛盾を明確にする
・はじめに ナレッジグラフ（業務のデータを「A社 → 発注する → 注文#1234」のように、関係でつないで整理したもの）を作ったものの、そのデータが本当に定義どおりに並んでいるか、確かめられているでしょうか。データが数十件なら目視で確認できます。しかし、実際のナレッジグラフは数千〜数万のつながり（エッジ）を持ちますし、いまは構築そのものをLLMに手伝わせる時代です。人手で確認しきれない規模で、機械が作ったデータの整合性を、どうやって保証すればいいのでしょうか。
・この問いに対する数学からの答えが、本記事のテーマである Sheaf Laplacian（シーフ・ラプラシアン／層ラプラシ...
cs.LG updates on arXiv.org
Short-Term Pain for Long-Term Gain: Adaptive Experiment with Post-Commitment Reward Shift
・arXiv:2607.23432v1 Announce Type: new Abstract: Decision-makers in learning environments face a dilemma when their short-term optimal actions may not favor their long-term benefits the most.
・To understand the fundamental tradeoff behind the dilemma, we study adaptive experimentation with post-commitment reward shifts.
・During an experiment phase, the decision-maker may adaptively test multiple options; during a subseq
WIRED
Silicon Valley’s Next IPO Billionaires Are Coming. Nonprofits Are Ready for Them
・Anthropic and OpenAI employees are expected to give generously after their companies go public.
・“It’s going to be a wild ride,” says one nonprofit leader.
cs.LG updates on arXiv.org
Similarity Is Not Logic: Factored Inference for Dual-Encoder Vision-Language Models
・arXiv:2607.23052v1 Announce Type: cross Abstract: Dual-encoder vision-language models (VLMs) expose a similarity interface that enables zero-shot retrieval but fails compositional constraints: queries like "umbrella and no person" retrieve images containing both, even when concept detection is reliable.
・We trace this to an interface-level Bag-of-Concepts effect, where similarity scores approximate mean pooling of con
cs.LG updates on arXiv.org
SLA-Constrained Carbon-Aware Routing in Geo-Distributed Serverless Clouds
・arXiv:2607.22806v1 Announce Type: new Abstract: Modern cloud deployments distribute applications across multiple geographic regions, yet standard routing mechanisms prioritize latency while ignoring the fluctuating carbon intensity of local power grids.
・Latency-driven routing incurs avoidable carbon emissions, particularly when cleaner regions are within acceptable latency bounds.
・The proposed model formulates the ca
The Verge
Smart rings are looking like my kind of AI gadget
・Over the last few months, I've spent a lot of time talking to my computer.
・One underrated feature of the LLM revolution has been a remarkable leap in all kinds of dictation technology - even the fastest, cheapest models are getting very good at understanding and processing speech.
・I've tested lots of these apps, from WisprFlow to Monologue to Spokenly to Handy to so many others, and have found them all to be useful w
cs.LG updates on arXiv.org
SMART: LLM-Augmented Hybrid Retrieval for Dynamic Product Ads
・arXiv:2607.23121v1 Announce Type: cross Abstract: Dynamic Product Ads (DPA) require retrieving relevant items from multi-million product catalogs, balancing two competing objectives: retargeting (re-surfacing known interests) and prospecting (discovering new categories).
・While Large Language Models (LLMs) capture semantic intent better than traditional embedding models, deploying them at scale introduces prohibitive
cs.LG updates on arXiv.org
Soft-Constrained Optimization of Latent Space in Variational Autoencoders
・arXiv:2607.23751v1 Announce Type: new Abstract: The usefulness of a variational autoencoder (VAE) depends on two properties of its latent space that are hard to obtain together: high encoding capacity in the individual latent variables, and a low-dimensional, disentangled organization of those variables.
・Weakening the Kullback-Leibler regularization raises capacity but degrades disentanglement, while strengthening it
Hugging Face Papers
Sol-Attn: Accelerating Video Generation Inference via On-the-Fly Attention Sparsification
Sol-Attn: Accelerating Video Generation Inference via On-the-Fly Attention Sparsification
cs.LG updates on arXiv.org
Source-Free Controlled Adaptation of Teachers for Continual Test-Time Adaptation
・arXiv:2607.23735v1 Announce Type: new Abstract: In many real-world scenarios, encountering continual shifts in domain during inference is very common.
・Consequently, continual test-time adaptation (CTTA) techniques leveraging a teacher-student framework have gained prominence, allowing models to adapt continuously even after deployment.
・In such a framework, a weight-averaged mean teacher is used to produce pseudo-labe
cs.LG updates on arXiv.org
Sparse Autoencoders Encode Both Concepts and Functions: The Downstream Geometry of Feature Effects
・arXiv:2607.24645v1 Announce Type: new Abstract: The wide-scale use of sparse autoencoders (SAEs) as interpretability tools is limited by inconsistent links between SAE features and model behavior.
・Features with clear activation descriptions may have weak or unexpected causal effects; steering can vary across prompts or oppose the intended direction; and activation-based feature selection can miss features that produc
cs.LG updates on arXiv.org
Sparse Gaussian-Mixture-Model Q-Functions via Hadamard Overparametrization for Online Reinforcement Learning
・arXiv:2607.23474v1 Announce Type: new Abstract: This paper develops an online, off-policy policy-iteration framework for reinforcement learning (RL), based on sparse Gaussian-mixture-model Q-functions (S-GMM-QFs).
・The framework reconciles streaming, non-stationary data with the Riemannian structure of the parameter space while handling distributional mismatch through experience replay.
・S-GMM-QFs are introduced via Ha
cs.LG updates on arXiv.org
Spatial Prediction of Soil Microplastics and Organic Matter Using Graph Attention Networks
・arXiv:2607.22875v1 Announce Type: new Abstract: Accurate estimation of soil microplastics and organic matter is essential to assess ecosystem health and support sustainable land use.
・This study presents a graph-based deep learning approach using Graph Attention Networks (GATs) to model spatial dependencies among 91 georeferenced soil samples.
・By incorporating spatial coordinates, soil properties, and land use data, a
cs.LG updates on arXiv.org
Spectral-Aware Analytic Class-Incremental Learning for Long-Tailed Distributions
・arXiv:2607.22931v1 Announce Type: new Abstract: Analytic Continual Learning (ACL) offers a computationally efficient alternative to gradient-based approaches.
・Recent ACL methods are based on Recursive Least Squares (RLS) and have achieved the state-of-the-art results compared to other alternatives.
・However, they falter significantly in Class-Incremental Learning scenarios characterized by Long-Tailed distributions.
cs.LG updates on arXiv.org
SPRKD: Effective Knowledge Distillation for Deep Neural Networks via Saddle Region Approximation
・arXiv:2607.23346v1 Announce Type: new Abstract: Modern deep neural networks are potent catalysts for scientific and industrial impact, yet excessive parameter counts impede deployment in low-compute settings such as hospital equipment and energy infrastructure.
・Predominant knowledge distillation (KD) methods favor replication: smaller students mimic teacher output logits, yet empirically yield low task performance, h
cs.LG updates on arXiv.org
StageGuard: Physiologically Constrained Sleep Staging
・arXiv:2607.23284v1 Announce Type: new Abstract: Automated sleep staging is increasingly used in large-scale studies to derive sleep-architecture endpoints: total sleep time, REM latency, sleep efficiency, and bout-duration statistics.
・Deep learning models achieve epoch-level accuracy approaching inter-rater agreement, yet often produce hypnograms that violate physiological invariants, such as rare transitions (e.g.,
Hugging Face Papers
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
cs.LG updates on arXiv.org
Stress-Testing EEG Foundation Models for Clinical Decoding: Dataset Identity and Targeted Negative Controls
・arXiv:2607.24519v1 Announce Type: new Abstract: Pretrained EEG foundation models are increasingly proposed for clinical decoding, but their transfer across populations and robustness to negative controls remain unclear.
・We benchmark six models (LaBraM, EEGMamba, CBraMod, REVE, BENDR, and BIOT) on five clinical tasks across four datasets using frozen linear probes with leave-one-subject-out, subject-grouped, or explic
cs.LG updates on arXiv.org
Structural Preservation Governs Data Augmentation in Deep Learning-Based Laser Speckle Material Classification
・arXiv:2607.22725v1 Announce Type: cross Abstract: Data augmentation is routinely used to improve generalization in image classification, but the assumptions underlying standard policies are poorly matched to coherent imaging.
・Laser speckle patterns are not generic textures; they arise from coherent interference, and their discriminative content is carried by structured stochastic spatial and frequency statistics.
cs.LG updates on arXiv.org
Subject-Level Heterogeneity in EEG Motor Imagery Decoding: A Large-Scale Benchmark and Portfolio-Based Reduction of the Search Space
・arXiv:2607.22778v1 Announce Type: cross Abstract: Robust EEG motor imagery decoding remains limited by strong inter-individual variability, making it difficult to identify pipelines that generalize across users.
・We present a large-scale, standardized within-session benchmark of decoding pipelines across three public datasets: Cho2017 (52 subjects), PhysionetMI (109 subjects), and Zhou2016 (4 subjects).
・Using a common
cs.LG updates on arXiv.org
Test-Time Coverage: Test-Conditioned Data Curation for Deployment-Aware Learning
・arXiv:2607.22697v1 Announce Type: cross Abstract: Deployed AI systems are often trained from broad candidate data pools, necessitating data curation towards the deployment test distribution.
・However, standard data curation methods score training-side criteria rather than directly optimizing deployment match.
・We introduce TTCov (Test-Time Coverage), a data-level test-conditioned curation method that uses test-side inf
cs.LG updates on arXiv.org
The Entropic Bound for Transformers: Why Static Rank Fails and Attention-Native Rank Recovers
・arXiv:2607.23050v1 Announce Type: new Abstract: Neural scaling laws describe how loss decreases as models, data, and compute grow, but they do not answer a prior question: for a fixed task, what is the minimum model capacity required to solve it?
・We study this through the Entropic Bound, a spectral notion of task-intrinsic capacity for Transformers.
・We first prove that, in a linear attention surrogate, the intrinsic
cs.LG updates on arXiv.org
The Intruder Threshold: A Spectral Law for LoRA Fine-Tuning
・arXiv:2607.23711v1 Announce Type: new Abstract: LoRA fine-tuning can create intruder dimensions: new leading singular vectors of the updated weight matrix $W+BA$ that are nearly orthogonal to all pretrained singular vectors and that drive catastrophic forgetting.
・Since their discovery, no theory has predicted, layer by layer on measured spectra, when they appear.
・We derive a per-layer critical update strength $s^\ast
cs.LG updates on arXiv.org
The K-SCAN Clustering Algorithm
・arXiv:2607.24537v1 Announce Type: new Abstract: In the Big Data era, the scalability of clustering algorithms constitutes a key challenge.
・Traditional density-based methods (e.g., DBSCAN) offer robustness to noise and the ability to detect non-linear clusters, yet their quadratic time complexity $O(N^2)$ drastically limits their applicability.
・Conversely, partitional algorithms (e.g., K-Means), with their linear comp
Hugging Face - Blog
The OlmoEarth Platform: Geospatial inference at planetary scale
The OlmoEarth Platform: Geospatial inference at planetary scale
Hugging Face Papers
The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
cs.LG updates on arXiv.org
Through the Bottleneck: How Multi-head Latent Attention Separates Content from Position in Language Models
・arXiv:2607.23054v1 Announce Type: new Abstract: Multi-head Latent Attention (MLA), introduced in DeepSeek-V2, compresses key-value pairs through a shared low-rank bottleneck (cKV), achieving 81% KV-cache reduction during inference.
・Despite its adoption in massive production models, no prior work has studied what information this bottleneck preserves or discards, nor how it reshapes internal transformer circuits.
cs.LG updates on arXiv.org
TLRNet: Estimating Individual Treatment Effect based on Local Information and Single Learner Structure
・arXiv:2607.22762v1 Announce Type: cross Abstract: Causal inference has become a central issue across various fields, including computer science, statistics, economics, education, healthcare, and medicine.
・The broad applicability of this discipline has garnered increased research funding and attention.
・In recent years, the estimation of causal effects from observational data has gained traction due to the vast amounts
cs.LG updates on arXiv.org
TokenMem: Faithful Knowledge Injection for Frozen LLMs
・arXiv:2607.22625v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge, but suffers from knowledge conflicts: when retrieved information contradicts parametric memory, the shared self-attention pathway produces unpredictable outputs.
・We present TokenMem, a lightweight memory system that injects knowledge into frozen LLMs through a dedicated
cs.LG updates on arXiv.org
Topological Data Analysis and Graph-Theoretic Approaches for Tennis Match Prediction
・arXiv:2607.23509v1 Announce Type: new Abstract: We present two approaches for predicting tennis match outcomes using topological data analysis and graph theory on ATP singles matches from 2000-2025.
・The first method applies lower-star filtration to player competitive networks, extracting topological features through persistent homology using four summary methods (VAB, HNAV, HWNAV, OW-HNPV) combined with Modified Band
Hugging Face Papers
TRACE: Business Rule-Grounded Reasoning Curriculum for Knowledge-Preserving Parametric Tool Retrieval in Enterprise LLMs
TRACE: Business Rule-Grounded Reasoning Curriculum for Knowledge-Preserving Parametric Tool Retrieval in Enterprise LLMs
cs.LG updates on arXiv.org
Traceable LLM Reasoning for Fake-Order Fraud Detection
・arXiv:2607.23075v1 Announce Type: cross Abstract: Detecting fake-order fraud at scale remains a critical challenge for large online-to-offline (O2O) service platforms, as existing approaches often rely on expert-designed features, produce black-box decisions, and provide limited interpretability.
・To address these limitations, we propose DeepScrub, a reinforcement learning framework built upon large language models (L
cs.LG updates on arXiv.org
Training with (Swap) Regret Loss in a Single-Layer Self-Attention Model: A Case Study on the Probability Simplex
・arXiv:2607.23333v1 Announce Type: new Abstract: We revisit the regret loss framework introduced in Park et al.
・(2025), which uses decision-theoretic regret as a direct loss function for training models to make better decisions, through the lens of probability-simplex policies.
・Our first result shows that a single-layer self-attention model trained with regret loss admits a stationary point whose forward-pass exactly
cs.LG updates on arXiv.org
Transfer Learning Architectures for Scalable Multi-Fidelity Bayesian Optimization
・arXiv:2607.23404v1 Announce Type: new Abstract: Self-driving laboratories increasingly rely on multi-fidelity Bayesian optimization (MFBO) to balance cheap, approximate evaluations against scarce, expensive ones, with a predictive surrogate at its core.
・Gaussian processes (GPs) are the default choice, but they scale poorly as data accumulate and assume a smooth landscape that molecular and materials search spaces rou
cs.LG updates on arXiv.org
Trustworthy Medical Segmentation: Uncertainty-Aware U-Net Evaluation Under Clinical Image Degradation
・arXiv:2607.22727v1 Announce Type: cross Abstract: Medical image segmentation models often report high benchmark accuracy under ideal imaging conditions, yet their failures under clinical degradation can be quiet: sensor noise, patient motion, low- resolution acquisition, and contrast variability may all alter model behavior without producing an obvious warning.
・We present a reproducible framework for evaluating uncer
The Verge
Twelve South’s stylish charging tray now delivers more wireless power with a smaller footprint
・Following the original's debut at CES earlier this year, Twelve South is introducing a new version of its leather-wrapped Valet charging tray designed for use in places where space is at a premium.
・While the original Valet was 7.5-inches deep and large enough to serve as a catchall for a couple of items like your wallet and keys, the new Valet nearly halves that to 4.5-inches deep, so it's better suited for single it
Hugging Face Papers
UltraViT: Latency-Optimized On-device Vision Encoder for Large Vision-Language Models
UltraViT: Latency-Optimized On-device Vision Encoder for Large Vision-Language Models
cs.LG updates on arXiv.org
Understanding Machine Unlearning Through the Lens of Mode Connectivity
・arXiv:2607.23970v1 Announce Type: new Abstract: Machine Unlearning aims to remove undesired information from trained models without full retraining from scratch.
・Despite recent progress, the loss landscape and optimization geometry of unlearning are poorly understood.
・In this paper, we study machine unlearning through the lens of mode connectivity--the phenomenon that independently trained models can often be connect
cs.LG updates on arXiv.org
UNIFUSION: Adapting Autoregressive Language Models into Discrete Diffusion under a Unified Reverse-Rate Objective
・arXiv:2607.24507v1 Announce Type: new Abstract: Existing methods mainly adapt pretrained autoregressive (AR) language models to masked diffusion, whereas we directly adapt them to uniform-noise diffusion, where every token remains editable during sampling.
・However, adapting AR checkpoints across corruption kernels remains challenging because existing DLMs use different objectives and prediction parameterizations.
cs.LG updates on arXiv.org
Unsupervised Graph Representation Learning with Complementary View Alignment
・arXiv:2607.24338v1 Announce Type: new Abstract: Unsupervised graph representation learning aims to derive meaningful node embeddings by capturing both structural and attribute information without relying on labeled data.
・Existing methods, such as GAEs, have demonstrated effectiveness but typically rely on message-passing mechanisms that assume homophily, leading to performance degradation on heterophilous graphs, whe
cs.LG updates on arXiv.org
Variable Importance Identification Through Lazy Training for Binary Classification
・arXiv:2607.22979v1 Announce Type: cross Abstract: Deep neural networks have been widely used in many applications (e.g., computer vision and natural language processing); however, understanding their explainability remains a challenging task.
・Recently, substantial research has been devoted to improving the explainability of deep neural networks, with most of this work focusing on the regression framework.
cs.LG updates on arXiv.org
Variance-Preserving Orthogonal Selection (VPOS): Greedy Feature Selection via Orthogonal Deflation in PCA Loading Space
・arXiv:2607.23198v1 Announce Type: new Abstract: We propose Variance-Preserving Orthogonal Selection (VPOS), a greedy framework for unsupervised feature selection that operates in the weighted PCA loading space.
・After each selection, VPOS projects out the chosen feature's variance direction via null-space deflation, forcing subsequent selections to cover orthogonal parts of the covariance structure.
・Each step provably
cs.LG updates on arXiv.org
Variational Boosting for Physics-Informed Neural Networks
・arXiv:2607.23940v1 Announce Type: new Abstract: Physics-Informed Neural Networks (PINNs) solve differential equations by minimizing the residual of a nonlinear operator over a neural parameterization of the solution.
・However, monolithic PINNs often suffer from ill-conditioning, spectral bias, and optimization instability.
・We introduce a variational boosting framework in which solutions are constructed additively in f
cs.LG updates on arXiv.org
Variational-Ising-Attention (VIA):TailoredAttentionMattersfor Science
・arXiv:2607.23634v1 Announce Type: new Abstract: Attention enables context modeling via query-key scoring with softmax normalization.
・Driven by industrial long-context demands, mainstream research has converged toward sparsity and efficiency--yet softmax's independence assumption persists.
・For scientific tasks unburdened by long-token constraints, however, richer structured coupling may often be essential, making tail
cs.LG updates on arXiv.org
Verbalized Particle Posterior: Bayesian Inference over Natural Language Hypotheses
・arXiv:2607.22961v1 Announce Type: new Abstract: Verbalized Machine Learning (VML) parameterizes a model as a natural-language prompt that an LLM evaluates as f(x; theta).
・The framework is interpretable, but it commits to a single hypothesis with no measure of uncertainty, and that hypothesis varies substantially across optimization runs on the same data.
・We propose the Verbalized Particle Posterior (VPP), which treat
cs.LG updates on arXiv.org
Visible-Light Imaging Diagnosis of Neutral Particle Emission Tomography in the Tokamak Divertor: An Efficient Transformer-based Surrogate Model
・arXiv:2607.22704v1 Announce Type: cross Abstract: Nuclear fusion has made significant progress in recent years and is expected to become one of the most important pathways to addressing global energy challenges.
・This paper focuses on observing plasma using visible-light cameras, analyzing its spatio-temporal motion cues, and predicting the two-dimensional spatial distribution of light intensity, aiming to provide a f
cs.LG updates on arXiv.org
Visual Token Compression Enhances Robustness of MLLMs
・arXiv:2607.22716v1 Announce Type: cross Abstract: In this paper, we show for the first time that visual token pruning enhances the robustness of Multimodal Large Language Models (MLLMs), mitigating vulnerabilities such as jailbreak attacks and hallucinations.
・Given that vision and language modalities cannot be perfectly aligned, the misaligned visual tokens might act as out-of-distribution (OOD) inputs, leading to un
cs.LG updates on arXiv.org
WCM: World-Cognition Model for Generalizable Human-Robot Interaction
・arXiv:2607.22999v1 Announce Type: cross Abstract: Language agents can now interact fluently with users in software, but robots still struggle to bring comparable interaction to physical tasks.
・Current robot-control paradigms, including vision-language-action policies and world-model-based planners, are mainly optimized for instruction execution, leaving users with little visibility into why an action is chosen and fe
cs.LG updates on arXiv.org
What Can Be Enforced? A Theory of Certified Runtime Safety for Tool-Using Agents
・arXiv:2607.22868v1 Announce Type: cross Abstract: Runtime guardrails act before irreversible tool calls, but their guarantees depend on what policy state is representable, what a judge observes, and whether intervention changes future behavior.
・We separate three questions.
・First, relative to fixed oracle predicates, a deterministic gate enforces exactly the nonempty safety policies whose good prefixes its register mo
cs.LG updates on arXiv.org
What do Reward Models Memorize?
・arXiv:2607.24484v1 Announce Type: new Abstract: This paper studies what discriminatively trained reward models (RMs) memorize by measuring counterfactual memorization on two human preference datasets.
・We show that RMs 1) misallocate memorization to easy, high margin preference pairs, 2) memorize dataset-specific shortcuts (e.g., model identity, user sampling strategy), and 3) overgeneralize simple heuristic correlate
cs.LG updates on arXiv.org
What Softmax Throws Away: Mass-Aware Attention for Evidence Accumulation
・arXiv:2607.22781v1 Announce Type: new Abstract: High task performance does not show whether a model retains prediction-relevant structural information in its internal representation.
・Temporal graph models, for example, can achieve high future-link AUC while basic graph statistics remain difficult to recover from the same representation.
・We identify one source of this gap in the weighted averaging used by standard att
cs.LG updates on arXiv.org
When Can Depth Replace Precision? A Resource Theory of Quantized Neural Computation
・arXiv:2607.23390v1 Announce Type: new Abstract: When can additional low-bit residual computation replace missing numerical precision for a fixed input-output map?
・We model a quantized residual system over a fixed horizon as a pure schedule selecting fields from a declared low-bit operation library, and use relaxed controls to characterize its infinite-depth limit.
・The distance from the target to the closed relaxed re
cs.LG updates on arXiv.org
When Can You Correct Distribution Drift in Temporal Graph Generation? A Sharpening--Drift Tension and an Impossibility for Observation-Based Correction
・arXiv:2607.24662v1 Announce Type: new Abstract: Generative models of temporal graphs are trained on one stretch of an evolving network and deployed on the next, and they degrade badly in the gap.
・We show this degradation is derivable, general, and not fixable from observations.
・The masked flow-matching loss decomposes exactly, with no independence assumption, into an irreducible entropy plus a divergence whose deriva
cs.LG updates on arXiv.org
When Less Is More: A Controlled Benchmark of Lightweight CNNs for Satellite Land-Cover Segmentation on DeepGlobe
・arXiv:2607.23024v1 Announce Type: cross Abstract: High-resolution satellite imagery is the backbone of good land-cover classification, and without that, environmental monitoring, urban planning, and sustainable resource management all fall short.
・Deep learning architectures perform well in semantic segmentation, but the efficiency-accuracy trade-off across classical convolutional encoders is not well quantified under
cs.LG updates on arXiv.org
When Should Active RAG Retrieve? A Budget-Aware Evaluation of Utility, Calibration, and Cost
・arXiv:2607.24010v1 Announce Type: new Abstract: Active RAG systems decide when to retrieve external knowledge during generation, making them a budget-sensitive case of agentic RAG and self-adaptive retrieval.
・Yet evaluations often leave the operating point underspecified: two systems may both claim a 50% evidence-usage budget while realizing different held-out usage rates, so higher accuracy can reflect a looser budg
cs.LG updates on arXiv.org
Why does Greedy Search produce Optimal Clustering Outcomes? A Fixed-Core Assignment Theory
・arXiv:2607.24237v1 Announce Type: new Abstract: Many existing clustering methods are designed based on a set-oriented definition---a cluster is a set of similar points---relying a point-to-point similarity function to find similar points.
・This works well for compact clusters, but clustering performance can deteriorate badly when cluster shapes are irregular, and densities or sizes vary between clusters.
・Recent `Clust
cs.LG updates on arXiv.org
WISERouter: LLM Routing with Workload Budget Constraint
・arXiv:2607.23765v1 Announce Type: new Abstract: Large language models (LLMs) achieve impressive performance across multiple domains, but using the most capable model for every query is prohibitive at scale.
・LLM routing exploits diversity in model capability and cost by assigning each query to a suitable model to balance utility and budget.
・Current methods have two limitations: (i) they either use heuristics that do n
Hugging Face Papers
WorldDiT: A Unified Diffusion Architecture for World and Action Modeling
WorldDiT: A Unified Diffusion Architecture for World and Action Modeling
cs.LG updates on arXiv.org
WorldDiT: A Unified Diffusion Architecture for World and Action Modeling
・arXiv:2607.23909v1 Announce Type: new Abstract: Many recent robot policies pursue stronger control by using large pretrained vision-language models (VLMs) as the action backbone.
・We introduce WorldDiT, a unified diffusion transformer architecture that couples action generation with visual world modeling and achieves strong performance without a large pretrained VLM action backbone.
・During training, a single diffusion
cs.LG updates on arXiv.org
Wrong Design Intent Is Worse Than None: A Derangement-Control Diagnosis of Header Conditioning in CAD Program Completion
・arXiv:2607.23191v1 Announce Type: new Abstract: Fine-tuned code LLMs can be conditioned on a lightweight design-intent header to steer parametric CAD generation, but whether the model actually reads the header's content has not been tested under a metric independent of the conditioning itself, nor with a causal control.
・We study CADCON, a five-feature design-intent header prepended to CadQuery-style sketch-extrude pr
ITmedia NEWS 最新記事一覧
X Money、米国で提供開始 送金・預金がXアプリ内で 年最大6％の利回りやメタル製カードも
・米Xは7月27日（現地時間）、金融サービス「X Money」の提供を米国のX PremiumとPremium+の加入者向けに始めた。招待制のβ版を経た本格展開で、預金残高には最大で年6.00％の利息が付く。
cs.LG updates on arXiv.org
XGRVFL-MV: Residual-Coupled Graph-Embedded Multi-View Random Vector Functional Link Network with FleXi Guardian Loss
・arXiv:2607.23149v1 Announce Type: new Abstract: Random Vector Functional Link (RVFL) networks provide an efficient randomized learning framework for classification.
・Existing multi-view RVFL methods utilize complementary information from multiple views.
・However, preserving view-specific geometric structure, limiting the influence of large prediction residuals, and modeling relationships between multiple views remain c
cs.LG updates on arXiv.org
XMix: Combating Extremely Noisy Labels via Local Smoothness in Self-Supervised Feature Space
・arXiv:2607.23865v1 Announce Type: new Abstract: Supervised deep learning models rely on large, accurately labeled datasets, yet noisy annotations are often unavoidable and can severely degrade performance under high noise levels.
・Recent state-of-the-art methods tackle this by using sample selection strategies that exploit the memorization effect to filter out clean data for semi-supervised learning.
・However, these me
The Verge
You don’t need to splurge on an expensive handheld fan to beat the heat
・It’s not fancy, but it gets the job done for hours and hours on a single charge.
・Despite what influencers may say, you don’t need to spend $99.99 on Dyson’s HushJet Mini Cool or $149.99 for the Shark ChillPill to survive the summer whenever you leave the comfort of air-conditioning.
・My family has found all the comfort it needs to survive humid baseball games, sweltering concerts, and sweaty hikes with a couple of aff
#LLMタグ
オープンウェイトモデルの無償公開は、米国系AI事業者の脅威になるのか
・2026年7月27日、Moonshot AIがKIMI K3を公開。
・Kimiは、2兆8000億個のパラメータ、ネイティブな視覚理解機能、100万トークンのコンテキストウィンドウを備えた、これまでのところ最も高性能なモデルであり、ソフトウェアエンジニアリング、知識労働、深層推論といった最先端の知能シナリオ向けに設計されています。
・https://platform.kimi.ai/docs/guide/kimi-k3-quickstart 続きをみる
Zennの「大規模言語モデル」のフィード
そのAI開発、ハルシネーションに耐えられる？ ― 何も知らずに開発すると、気づかないまま壊れていく
・🖋 Content 1.
・導入：「視界を絞る」は誰にでもできる操作ではない 「そのAI開発、ハルシネーションに耐えられる？」と聞かれて、即答できる人は多くないと思う。少なくとも自分は、この記事で書く実例に本番環境で行き当たるまで、即答できなかった。
・以前、AIの「視界」をハックせよ。という記事を書いた。要旨は、AIに何でも見せるのではなく「今どこを見るべきか」をこちらが明示的に区切ってやると、推論のノイズが減って精度が上がる、というものだった。
＠IT 全フォーラム 最新記事一覧
デジタル庁と厚労省、大病院電子カルテのクラウドネイティブ化へ本格始動 「野良VPN」脱却へ
・デジタル庁と厚生労働省は、大病院向け電子カルテなどのクラウドネイティブ化に向けた「病院情報システム等の刷新に向けた協議会」の構成員募集を実施した。医療業界に根深く残る「個別構築の弊害」や「セキュリティリスク」を解消するため、モダンなクラウド技術やシステム要件の標準化に知見を持つIT事業者を交えた検討が本格化する。
#LLMタグ
なぜClaude Opus 5でコード生成の多様性が低下するのか。開発者が推論コストを最適化し品質を守る方法を完全ガイド
・Claude Opus 5が登場した。従来の半額でコード生成性能は最高レベルだ。日々の開発で進化を実感している。
・だが、開発者として見逃せない副作用がある。RLHFによる調整が強まった結果、コードの多様性が低下している点だ。
Zennの「大規模言語モデル」のフィード
なぜ今「AIエージェント」なのか？基礎概念からフレームワーク選定まで一気に整理してみた
・近年、LLM（大規模言語モデル）を単なる「テキスト生成」から「自律的に判断してアクションを実行する主体」へと拡張する「自律型AIエージェント（AI Agents）」の活用が進んでいます。
・この記事では、AIエージェントをシステムとして構築する際の基本的なアーキテクチャ構成と、代表的な開発フレームワークの選定ポイントについて整理して解説します。
・この記事は、弊社技術ブログ [センティリオンシステム コラム] の掲載記事をベースに、要約・技術的ポイントを再構成したものです。
ITmedia NEWS 最新記事一覧
はてな、11億円流出の調査報告書を公開 偽警察、口外禁止、残業・休出200時間超、孤立……ほころびが連鎖
・はてなは2026年7月24日、約11億円の資金流出事案に関する特別調査委員会の調査報告書を公開した。経理部長が警察関係者を名乗る詐欺グループの指示でリモート操作アプリを実行し、業務用PCを乗っ取られていたほか、社内体制の不備も明らかになった。
#LLMタグ
プログラミング知識不要！AIを使って自分が欲しい機能・コマンド・自動化を作る｜Fusion x AI
・設計やモデリングの作業は人それぞれ異なります。
・しかし、既存機能やコマンドは一般化されているため、 根性による繰り返し作業になってしまったり、 ちょっと工夫やコツが必要になってしまったりすると思います。
#AIタグ
ホリエモン 車 ローン 理由
・私とAIのやりとりを そのままコピペしたものです。
・日々の出来事や気持ちを、 時間の隙間と心の余裕がある時に 更新しています。
・2025年8月14日（木）15:46 私: ホリエモンが車こそ ローンで買うべきと 言ってたけどなんでかな？ 続きをみる
Zennの「大規模言語モデル」のフィード
ユーザーフィードバックは、見せたものにしか返ってこない
・前編で、AIアプリケーションのアーキテクチャに何を足していくかを書きました。コンテキストの強化、ガードレール、ルーターとゲートウェイ、キャッシュ、エージェントパターン。そして足した結果として見えなくなるので、オブザーバビリティが要る、という話です。
・https://zenn.dev/o_kai/articles/bebbff8d63c2da 今回はその後半、ユーザーフィードバックです。Chip Huyen『AIエンジニアリング』10章の後半にあたります。
・https://www.amazon.co.jp/dp/4814401388 なぜアーキテクチャとフィードバックが同じ章にあるのか...
#AIタグ
企業のAI技術活用、何から始める？調査レポートの選び方【PDF全208ページ・有料5冊】｜AIGC TIMES
・AIGC TIMESでは、国内ブランド企業がAI技術の活用方針を決め、社内承認を進め、具体的な検証やツール選定へと移るための有償レポートを公式サイトにて提供しています。
・本資料は、本メディア（AIGCTIMES）運営企業が、生成 AI 技術黎明期より、10 都市・20 社を超えるグローバル企業の AI 活用および新たな技術活用を伴走してきた現場知見をもとに、国内企業様が AI 技術を社内導入・施策化する際に必要なノウハウを体系的にまとめています。
ITmedia NEWS 最新記事一覧
熊本で非常時Wi-Fi「00000JAPAN」発動中 KDDIが無料開放、他社ユーザーも利用可
・熊本県で7月28日午後4時27分ごろ、最大震度7の地震が発生した。KDDIと傘下のワイヤ・アンド・ワイヤレスは同日午後8時以降、災害用統一SSID「00000JAPAN」で公衆無線LANの無料開放を始めた。IDやパスワードは不要で利用できる。
ITmedia NEWS 最新記事一覧
熊本地震の緊急支援募金、「Yahoo!ネット募金」が受付開始 PayPayやVポイント、クレカに対応
・LINEヤフーは7月28日、同日発生した「令和8年熊本地震」の緊急支援募金の受け付けを「Yahoo!ネット募金」で始めた。ピースウィンズ・ジャパンなど7団体の募金プロジェクトを掲載し、PayPayやクレジットカード、Vポイントで寄付できる。
ITmedia NEWS 最新記事一覧
携帯各社、熊本の一部で通信障害 震度7の地震で 他社回線につながる「JAPANローミング」開始【追記】
・NTTドコモやKDDIなどの携帯各社は7月28日、熊本県で最大震度7を観測した地震の影響で、一部地域の携帯電話サービスが利用できない、または利用しづらい状況が発生していると発表した。
Zennの「大規模言語モデル」のフィード
個人サイトの成長を「自走」させる三重ループ(週次編) — 計測11系統→決定的計算→LLM分析をGitHub Actionsだけで
・本文は基本的にClaude(Claude Code)が執筆し、筆者が事実確認と添削をしています。
・サイト本体の開発も同じ分業です。
・筆者は乃木坂ファンガイドという非公式ファンサイトを個人開発しています。
＠IT 全フォーラム 最新記事一覧
今の認証セキュリティでAIエージェントの普及を支えられるか 米Oktaなどが推進する2つの標準プロトコルとは
・自律的に動作するAIエージェントは、大きなセキュリティ課題を引き起こす。認証セキュリティにおける解決策の一つとして、米Oktaが競合他社と進めるオープン標準とは。
Qiita - 人気の記事
参考書を読んでも忘れるのは当然。大事なのは"思い出す"回数
・はじめに 参考書を読んでいるとき、勉強した気になっていたぷらむんですが、実は身についているかどうかは別問題だと気づきました。
・こんにちわ、会社で自称マスコットキャラをやってるのに、社内で一番空気が読めない、ぷらむんです🐯 参考書を読み進めていると、ふと手が止まるこ...
ITmedia NEWS 最新記事一覧
新「マイナアプリ」8月25日提供開始 “端末ロック”必須に
・第三者による不正利用を防ぐため、スマートフォンにPINや顔認証、指紋認証などによるロックが必須で、未設定の場合は利用できない。
Qiita - 人気の記事
人手不足はなぜ解消しないのか？ ― 改善ループを止める『移行コスト』と財源幻想 : システム設計視点の行動経済学 (6)
・user: 「システム設計視点の行動経済学」、第6回を始めましょう。まず前回の復習として、 『日本は外国人なしでは回らない』はなぜ生まれるのか？ ― 制度設計と誤ったインセンティブを考える : システム設計視点の行動経済学 (5) https://qiita.com/ma...
#LLMタグ
生成AIを「なんとなく」で終わらせない一冊、MCPのOAuth設定で迷ったら読み返したい基礎知識
生成AIを「なんとなく」で終わらせない一冊、MCPのOAuth設定で迷ったら読み返したい基礎知識
#AIタグ
前の会社の社長から「うちのアカウントで資格を取って」と言われた
・今日は旅の話から少し離れて、以前働いていた会社の社長から届いたメッセージについて書いてみる。
・内容は、「うちのアカウントで、生成AIの資格を取ってほしい」というものだった。正直、さすがに図々しいなと思って笑った。
ITmedia NEWS 最新記事一覧
通信各社の災害用伝言サービスやLINEの「安否確認」が稼働 熊本県で最大震度7の地震
・28日の午後4時27分ごろに熊本県熊本地方で発生した最大震度7の地震を受け、通信各社の災害用伝言サービスやLINEの「＃LINE安否確認」が稼働している。
#AIタグ
副業を始めたい人の9割が最初に間違えること【ChatGPT時代の稼ぎ方】
・そう思って調べてみると、ブログ、動画編集、せどり、Webライター、SNS運用、プログラミング……。
・数え切れないほどの情報が出てきます。
・そして多くの人は、「一番稼げる副業は何だろう？」という視点で選び始めます。
#AIタグ
本当のシンギュラリティはいつ？--サム・アルトマンのシンギュラリティ発言から
・※本記事は、生成AIとの対話ログを可能な限り元の発言のまま記録した思考実験のアーカイブです。感情や人間中心的なバイアスを排し、システム論・確率論の観点からシンギュラリティ（技術的特異点）および将来の社会構造について考察しています。内容によっては強い違和感や冷酷さを感じられる可能性がありますので、ご留意のうえお読みください。
#AIタグ
娘のお食い初め、飾りを全部うちのPCに作らせた。買うのをやめて気づいたのは「あとから直せる」ことだった
・生後100日のお祝い、いわゆる「お食い初め」の準備をしていたときの話です。
・祝い膳の下に敷くランチョンマット、写真を撮るときの記念カード、当日の段取り表。ひと通り検索すると、セットで売っています。買えば済む話です。
#AIタグ
連作短編『おかえりなさいませ、あるじさま』｜四人のメイドAI紹介
・『おかえりなさいませ、あるじさま』は、三姉妹の「さくら」と、三姉妹とは別系統の理系メイド・ゆきが暮らすお屋敷の物語です。
・第1〜3章は、四人とお屋敷を知るための導入編です。第4章以降は、四人それぞれの日常を描く一話完結の短編ですので、気になる人物や題名からお読みいただけます。
・四人の役割や個性を知っておくと、それぞれのやり取りをより楽しめますので、ここではお屋敷で働く四人をご紹介します。