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。
cs.LG updates on arXiv.org
A Cascaded Edge-Cloud Architecture for Automated Diabetic Retinopathy Screening
・arXiv:2605.14108v2 Announce Type: replace-cross Abstract: Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness, and automated screening can help extend specialist capacity in resource-constrained clinical workflows.
・Cloud-based deep learning systems can provide strong grading performance, but they require image upload and reliable connectivity.
・We evaluate a two-tier edge-cloud cascade on t
ITmedia NEWS 最新記事一覧
BYDが作った軽EV「RACCO」発売 214万5000円から CEV補助金込みなら200万円切り
・BYD Auto Japanは、日本の軽自動車規格向けに専用設計したEV「BYD RACCO」を発売した。価格は214万5000円から。航続距離は最大320kmで、OTA更新や長期保証にも対応する。
ITmedia NEWS 最新記事一覧
TSMC熊本工場、段階的に通常操業へ 熊本の地震で一時中断、従業員の無事確認 台湾報道
・半導体大手の台湾TSMCは7月28日夜、同日午後の最大震度7の地震で操業を一時中断した子会社JASM（熊本県菊陽町）について、従業員全員の無事を確認し、段階的に通常操業に戻りつつあると回答した。台湾紙・経済日報が報じた。
#AIタグ
「もう、プログラミングはいらない。」
・Claude Codeを触るようになってから、毎日のように新しいものを作っていました。
・ある日、ふと思ったことがあります。
・「面白いスロットゲームを作ってみたい。」 でも、普通のスロットじゃ面白くありません。
#AIタグ
【無料公開】ChatGPTを使っているのに成果が出ない人へ
・「毎日使ってるのに、結局何も変わらない。」 もし、あなたがこんな風に感じたことがあるなら、この記事はきっと役に立ちます。
・ChatGPTに質問しても、どこか物足りない回答しか返ってこない。
・SNSでは「AIだけで月○万円稼げました」という投稿を見かけるのに、自分は何も成果が出ない。
#AIタグ
AIで、一番助けたかった人。
・ラーメン屋のツールを作ってから、僕は毎日のように考えていました。
・「次は何を作ろう。」 AIを触っていると、不思議とアイデアが次々と浮かんできます。
・その中で、真っ先に思い浮かんだ人がいました。
cs.LG updates on arXiv.org
Predicting Channel Closures in the Lightning Network with Machine Learning
・arXiv:2605.12759v2 Announce Type: replace Abstract: The Lightning Network (LN) is a second-layer protocol for Bitcoin designed to enable fast and cost-efficient off-chain transactions.
・Channels in the LN can be closed either by mutual agreement or unilaterally through a forced closure, which locks the involved capital for an extended period and degrades network reliability.
・In this paper, we study the problem of pred
ITmedia NEWS 最新記事一覧
ソニー、熊本の半導体工場で生産停止 令和8年熊本地震で 建屋などの被害状況を確認中
・ソニーセミコンダクタソリューションズは7月29日、28日に発生した「令和8年熊本地震」による生産拠点への影響を発表した。グループ会社の本社兼生産拠点「熊本テクノロジーセンター」（熊本県菊陽町）では、地震発生直後から生産を停止している。
#AIタグ
「AIなら簡単に稼げる」と思っていた。でも現実は違った。
・最近は、AIと一緒に趣味のサイトを作りながら、収益化にも挑戦しています。
・記事の構成やSEO、サイト設計までAIに相談しながら進めていますが、実際にやってみて分かったことがあります。
#AIタグ
「コンテンツ？欲しけりゃくれてやる。探せ！この世のコンテンツを全てをそこにおいてきた！」世は正に大コンテンツ時代！
・ありったけのコンテンツかき集め 毎日毎日、大量の情報が押し寄せて参ります。
・もちろん例に漏れず、私も毎日毎日、大量の情報の波に飲まれるのです。
#LLMタグ
「ゼロから作るDeep Learning6 LLM編」
・ゼロつくは全作読んでいるが待望のLLM編。
・LLMに使われる基本的な内容を学んだ上で、GPT2を自分で実装してみようというセッションがあり、自ら手を動かし実装することで理解が深まった。
・次の章では、作成した基本的なモデルを効率的に、早く動かすにはどうすればいいか紹介され、実際に実装すると「確かに早い！」と体験することができ面白い。
#AIタグ
「今、家の中で起きているAI事件簿」〜母ちゃんと息子の家庭内捜査〜
・相(AI)棒と共に、日々の経験や気づきをnoteにしながら、自分なりのライフワークを育てている、高井 道です。
・いつものなんとなく難しく見える出来事や情報などを、身近に感じられる日常の物語りに自分なりの視点で変換したノンフィクション作品です。
#AIタグ
「災害参謀AI」― まだ骨組みだけの話
・昨日、2016年熊本地震の教訓をもとに、避難所ごとの物資と人数を記録する簡単なツールを作った記事を書いた。
・一晩経って考えたのは、「物資だけでなく、避難所の状況そのものを、もっと早く・楽に報告できないか」ということだった。
・テキストで入力するのは、混乱した現場では意外と手間がかかる。
ITmedia NEWS 最新記事一覧
「受信料の値上げは検討していない」NHK井上会長が明言 策定中の3カ年経営計画巡り
・NHKの井上樹彦会長が「現時点で受信料の値上げは検討していない」と述べた。井上氏は「受信料の水準も計画の中で示すことになっているが、あくまでも経営の合理化と、受信料外収入の確保などの努力を尽くしていく」とした。
#LLMタグ
「生成AIが〜🤪」と語る時代遅れな知識人たちへ
・いつの時代も、変化の最前線から一歩引いた安全な場所から、知的なポーズを取って冷笑する人々がいる。
・かつてインターネットやスマートフォンが登場したときもそうだった。
＠IT 全フォーラム 最新記事一覧
「年収2000万円以上が約6割」 CTO、CIOなどハイクラス転職が5年で2.2倍のバブル、求められる人材像は
・クライス＆カンパニーがCXO・経営幹部人材の転職市場動向を分析した「CXO転職市場レポート2026」を公開した。高年収帯のCXO採用が増加した他、スタートアップ企業比率の増加や高水準の非公開求人比率から、経営人材市場の流動性がさらに高まっているという。
LLMタグが付けられた新着記事 - Qiita
【2026年7月】Claude Opus 5がリリース！Fable 5級の性能を半額で・API変更点まとめ
・はじめに 2026年7月24日、Anthropicから Claude Opus 5 が一般提供開始されました。2026年6月にリリースされた Fable 5 からわずか1ヶ月余りでの新モデル投入で、「Fable 5のフロンティア級知性に迫る性能を、半額のコストで実現した」...
#LLMタグ
【AI開発のリアル #20】 ローカルLLMの仕事は、失敗から決めた
・前回、LLMは、入力のたびに別の応援の人が来るようなものだと書いた。
・前の仕事を覚えていないから、毎回、台本と必要な資料を渡す。その性質に合わせ、専門家の仕事も、一つずつ独立して渡せる形へ分けた。
#AIタグ
【AI活用中級編】AIに社則を守らせる
・社員全員がAIの投資会社、ツキヨミ・キャピタルの技術解説・中級編です。AIに仕事を頼むと、最初はうまくいくのに、回数を重ねるとだんだん言うことを聞かなくなる。あの現象への対処が「社則を書いて、毎回読ませる」でした。うちで実際に守られなかった社則と、破られたあとに何をしたかまで、正直に公開します。
Qiita - 人気の記事
【AWS勉強中】新卒1年目が「さくらのAI Engine」でAWS問題が出てくるガチャを作ってみた
・はじめに みなさんこんにちは！！ 今年からエンジニアになり、AWSの勉強を始めました！ AWS Certified Cloud Practitionerは取得できたのですが、時間が経つと知識が離れていく...
・弊社では毎日お昼にAWSの問題を解いているのですが、新人の私か...
Zennの「大規模言語モデル」のフィード
【Claude Code】SKILLを作ってインストールする
・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。
・C1でSKILL.mdの構造を理解しました。次は自分でSKILLを作ってみましょう。
・この記事では、実際にzenn-article-generatorというSKILLをゼロから作った体験を軸に、SKILL.mdの書き方・ディレクトリの作り方・Claude Codeに認識させるまでの流れを解説します。
LLMタグが付けられた新着記事 - Qiita
【LLM構造化出力】ラズパイ自作ロボットを自然言語対話で制御してみた
・第1章：導入（自然言語でロボットを安全に動かしたい！） はじめに：この記事を書いたきっかけ（開発の経緯） この記事の執筆に至ったのは、趣味のラズパイ工作がきっかけです。
・Raspberry Pi と PiCar-X のキットを使って自作ロボットを作り、しばらく動かしてい...
#LLMタグ
【ローカルLLM】Macaron-V1-Tallを使ってみた話【Qwen3.6 35B A3B】
【ローカルLLM】Macaron-V1-Tallを使ってみた話【Qwen3.6 35B A3B】
機械学習タグが付けられた新着記事 - Qiita
【技術解説】【完全ガイド】XGBoostを用いた特徴量重要度の算出と株価予測の実装
・【完全ガイド】XGBoostを用いた特徴量重要度の算出と株価予測の実装 本記事では、PythonのライブラリであるXGBoostを活用し、株価予測のための特徴量重要度の算出手法を解説します。これにより、データサイエンスを活用した投資戦略の構築に役立てることができます。
#LLMタグ
【雑記】AIが無数にいるせいで、「私は私だけ」が怪しくなっていく
・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。
・AIと話していて、よくよく考えたらこわぁ！と思うことがあります。
・同じカスタム指示で複数のチャットを行う時、両方に同じ相手がいますよね。
#LLMタグ
【生成AIニュース+】『GPT-Live-Transcribe / GPT-Transcribe』『PrunaVAED』『Fish Audio S2.1 Pro』『LTX-2.3-22b-IC-LoRA-LensRemover』『A.X-K2』『text-to-cad』『Mirage Avatar X』『ComfyUI-muscriptor』『Hanji』『Krea2 depth lrzjason』『Rook』『ヒューマノイド掃除サービス』『AGIBOT A3』
【生成AIニュース+】『GPT-Live-Transcribe / GPT-Transcribe』『PrunaVAED』『Fish Audio S2.1 Pro』『LTX-2.3-22b-IC-LoRA-LensRemover』『A.X-K2』『text-to-cad』『Mirage Avatar X』『ComfyUI-muscriptor』『Hanji』『Krea2 depth lrzjason』『Rook』『ヒューマノイド掃除サービス』『AGIBOT A3』
#AIタグ
【日記】今日のロン毛とか、YESロン毛・NOタッチの話、AIっぽくなった不思議など
・夜中に目が覚めてしまったので日記と、今日のラクガキまとめなど・・・ （ミッドナイトに書いているのでなんか恥ずかしくなったら消すかもしれません、が、こういうのは大抵消さない） ーーー 続きをみる
Zennの「大規模言語モデル」のフィード
☕ローカルLLMを回したままMacのフタを閉じたくて、スリープを防ぐmacOSアプリを個人開発した☕
・きっかけ：フタを閉じるとローカルLLMが止まる Mac（Apple Silicon）でclaudeCode・ローカルLLMを回すのが当たり前になってきました。
・Ollama や LM Studio、MLX で長時間の推論・学習・バッチ処理を流す。
・ところが困るのが、 **「実行中にMacBookのフタを閉じると強制スリープしてジョブが止まる」**問題です。
#AIタグ
2026年7月の読んだ本
・初めまして、あんなかと申します。
・noteでは初めての記事になりますが、私、はてなブログで温泉ライターもやっておりました。
・鳴かず飛ばずでしたが… youtube、X もやっておりますのでよろしければフォローよろしくお願いいたします。
WIRED
23 WIRED-Approved Gifts for Frequent Travelers (2026)
・For the frequent flier who treats the Delta Sky Club like their second home.
#AIタグ
7日で11,546再生いったYouTubeを、AIに「直せない」と言われて全部消しました
・7日間で11,546回再生されたYouTubeチャンネルを、私は自分の手で全部削除しました。
・きっかけは、制作を手伝ってもらっていたAIが「このバグは直せない」と言ったことです。
#LLMタグ
A Canonical Interpretation of NRA-IDE
・Core Axiom, Boundary Structure, and Authority Separation NRA-IDE Canonical Interpretation Series — Part I of III 続きをみる
cs.LG updates on arXiv.org
A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
・arXiv:2607.23806v1 Announce Type: cross Abstract: Improving a language model today means retraining it: enormous compute, a new opaque model each cycle, non-deterministic output.
・We take the opposite path: the model stays frozen, and a persistent memory of verified solutions grows beside it.
・Once a problem family is solved and has passed an independent verification step that never consults the answer key, every new i
cs.LG updates on arXiv.org
A General Deep Learning Framework for Wireless Resource Allocation under Discrete Constraints
・arXiv:2603.19322v2 Announce Type: replace Abstract: While deep learning (DL)-based methods have achieved remarkable success in continuous wireless resource allocation, efficient solutions for problems involving discrete variables remain challenging.
・This is primarily due to the zero-gradient issue in backpropagation, the difficulty of enforcing intricate constraints with discrete variables, and the inability in gener
stat.ML updates on arXiv.org
A Generalized Tangent Approximation based Variational Inference Framework for Strongly Super-Gaussian Likelihoods
・arXiv:2504.05431v3 Announce Type: cross Abstract: Variational inference, as an alternative to Markov chain Monte Carlo sampling, has played a transformative role in enabling scalable computation for complex Bayesian models.
・Nevertheless, existing approaches often depend on either rigid model-specific formulations or stochastic black-box optimization routines.
・Tangent approximation is a principled class of structured
cs.LG updates on arXiv.org
A Mechanistic Perspective and Circuit-Guided Difficulty Metric for Unlearning
・arXiv:2601.09624v2 Announce Type: replace Abstract: Machine unlearning is becoming essential for building trustworthy and compliant language models.
・Yet unlearning success varies considerably across individual samples: some are reliably erased, while others persist despite the same procedure.
・We argue that this disparity is not only a data-side phenomenon, but also reflects model-internal mechanisms that encode and p
cs.LG updates on arXiv.org
A Model for Imbalanced Label Aggregation: A Focus on Minority-Class Detection
・arXiv:2607.24622v1 Announce Type: cross Abstract: We study imbalanced crowdsourcing with a focus on class-dependent annotator accuracy, a setting that, to the best of our knowledge, remains relatively underexplored despite its importance in real-world inspection systems where the labels of greatest operational importance are also the rarest ones.
・In this setting, annotators may be reliable on both classes, unreliable
Hugging Face Papers
A New Role for Relevance: Guiding Corpus Interaction in Agentic Search
A New Role for Relevance: Guiding Corpus Interaction in Agentic Search
cs.LG updates on arXiv.org
A Simple, Optimal and Efficient Algorithm for Online Exp-Concave Optimization
・arXiv:2512.23190v3 Announce Type: replace Abstract: Online eXp-concave Optimization (OXO) is a fundamental problem in online learning, where the goal is to minimize regret when loss functions are exponentially concave.
・The standard algorithm, Online Newton Step (ONS), guarantees an optimal $O(d \log T)$ regret, where $d$ is the dimension and $T$ is the time horizon.
・Despite its simplicity, ONS may face a computationa
stat.ML updates on arXiv.org
A Support-Set Algorithm for Optimization Problems with Nonnegative and Orthogonal Constraints
・arXiv:2511.03443v2 Announce Type: replace-cross Abstract: In this paper, we investigate optimization problems with nonnegative and orthogonal constraints, where any feasible matrix of size $n \times p$ exhibits a sparsity pattern such that each row accommodates at most one nonzero entry.
・Our analysis demonstrates that, by fixing the support set, the global solution of the minimization subproblem for the proximal line
cs.LG updates on arXiv.org
A Survey of Graph Transformers: Architectures, Theories and Applications
・arXiv:2502.16533v3 Announce Type: replace Abstract: Graph Transformers (GTs) have demonstrated a strong capability in modeling graph structures by addressing the intrinsic limitations of graph neural networks (GNNs), such as over-smoothing and over-squashing.
・Recent studies have proposed diverse architectures, enhanced explainability, and practical applications for Graph Transformers.
・In light of these rapid developm
cs.LG updates on arXiv.org
A2QTGN: Adaptive Amplitude Quantum-Integrated Temporal Graph Network for Dynamic Link Prediction
・arXiv:2605.21916v2 Announce Type: replace-cross Abstract: Dynamic link prediction is important for modeling evolving interactions in social, communication, financial, and transportation networks.
・Classical temporal graph models capture changes over time, but they may struggle to represent rapidly evolving node-edge interactions in large dynamic graphs.
・We propose A2QTGN (Adaptive Amplitude Quantum-Integrated Temporal
cs.LG updates on arXiv.org
Accelerating Frequency Domain Diffusion Models with Error-Feedback Event-Driven Caching
・arXiv:2604.22901v2 Announce Type: replace Abstract: Diffusion models achieve remarkable success in time series generation.
・However, slow inference limits their practical deployment.
・We propose E$^2$-CRF (Error-Feedback Event-Driven Cumulative Residual Feature caching) to accelerate frequency domain diffusion models.
cs.LG updates on arXiv.org
Accelerating Hierarchical Sparse Predictive Coding with Hybrid Amortized Inference
・arXiv:2606.27802v2 Announce Type: replace Abstract: Hierarchical predictive coding provides an interpretable framework for perception as error-driven inference in multi-layer models, while sparse coding imposes parsimonious latent representations through explicit sparsity constraints.
・Their combination yields hierarchical sparse predictive coding models with appealing computational and neuroscientific properties, but
OpenAI News
Accelerating scientific discovery with ChatGPT for Academic Researchers
・OpenAI is giving 100,000 academic researchers free access to ChatGPT's most advanced AI models to accelerate scientific research, collaboration, and discovery.
cs.LG updates on arXiv.org
Action from Adjacent Set in Physical Space Outperforms the Best Prediction in World Models
・arXiv:2607.23602v1 Announce Type: cross Abstract: Controllers based on sampling and latent world models assign a predicted terminal cost to each candidate action sequence, choose the minimum, execute its first action block, and replan.
・This rule can fail even when the terminal cost perfectly and accurately reflects the true task objective in the physical world.
・Residual prediction error can give an infeasible sequenc
cs.LG updates on arXiv.org
Action-Sufficient Goal Representations
・arXiv:2601.22496v2 Announce Type: replace Abstract: In offline goal-conditioned reinforcement learning (GCRL), hierarchical approaches decompose long-horizon tasks into high-level subgoal prediction and low-level action execution.
・A critical design choice in such architectures is the goal representation-the compressed encoding of goals that serves as the interface between these levels.
・Existing methods derive this re
cs.LG updates on arXiv.org
Additive Control Variates Dominate Self-Normalisation in Off-Policy Evaluation
・arXiv:2602.14914v3 Announce Type: replace Abstract: Off-policy evaluation (OPE) is essential for assessing ranking and recommendation systems without costly online interventions.
・Self-Normalised Inverse Propensity Scoring (SNIPS) is a standard tool for variance reduction in OPE, leveraging a multiplicative control variate.
・Recent advances in off-policy learning suggest that additive control variates (baseline correct
Hugging Face Papers
Agent Retrieval Bench: Evaluating Repository Context Retrieval for Coding Agents
Agent Retrieval Bench: Evaluating Repository Context Retrieval for Coding Agents
cs.LG updates on arXiv.org
Agent-UCT: Upper Confidence Bounds Applied to Trees for Agentic Workflow Optimization with Cost-Awareness
・arXiv:2607.24162v1 Announce Type: cross Abstract: Optimizing agentic workflows, such as retrieval-augmented generation (RAG) pipelines, requires navigating a combinatorial space of discrete component choices under tight evaluation budgets.
・Existing approaches - heuristic search, black-box optimization, and standard tree search methods - do not explicitly exploit the compositional structure of these workflows, leading
cs.LG updates on arXiv.org
AI Strategy: How to Choose What AI Product to Implement
・arXiv:2607.23733v1 Announce Type: cross Abstract: Firms struggle to choose AI projects that pay off: two projects can look equally promising to smart, motivated stakeholders and yet deserve opposite decisions.
・At the residential real-estate brokerage Compass, one AI product (Likely-to-Sell recommendations) flagged sales outreach opportunities and went on to account for nine figures in annual gross commission revenue.
#LLMタグ
AI オープンソース、オープンウェイト、KimiK3ついて書いてみた
・「AIに質問すると、人間以上に丁重で、いつも前向きな返事が返ってくる」。
・この現象の裏側には、人間のような意思を持つ魔法の脳みそが存在するわけではありません。
・その実態は、入力された言葉を高速で数値化し、確率計算を繰り返しながら、最終的には「太鼓持ち」のような役割のプログラムが完璧に味付けをして出力を行うという、巨大な自動システムです。
cs.LG updates on arXiv.org
AIFL: A Global Daily Streamflow Forecasting Model Using a Deterministic LSTM Pre-trained on ERA5-Land and Fine-tuned on IFS
・arXiv:2602.16579v2 Announce Type: replace Abstract: Reliable global streamflow forecasting is essential for flood preparedness and water resource management, yet data-driven models often suffer from a performance gap when transitioning from historical reanalysis to operational forecast products.
・This paper introduces AIFL (Artificial Intelligence for Floods), a deterministic LSTM-based model designed for global daily
#AIタグ
AIエージェントの3つの弱点を消すMCPサーバー活用術 codebase-memory-mcp・cognee・OpenRouterを組み合わせる
・Claude CodeやCursorでAIエージェントに作業を任せていると、こんな場面に何度も出会いませんか。
・前日にどこまで進めたかを説明しても、今日のセッションでは最初から覚え直しになる。
・大きめのリポジトリに手を入れてもらうと、目的のファイルにたどり着くまで何度も探索を繰り返す。
機械学習タグが付けられた新着記事 - Qiita
AIエンジニアとして本番環境で戦うための書籍5冊 ― 全体像から実装・運用まで
・はじめに 自分がAIエンジニアリングの領域に本格的に足を踏み入れたのは2024年の後半だった。それまではバックエンド寄りの開発をしていたが、社内でLLMを使ったプロダクトを立ち上げることになり、否応なしにこの世界へ引き込まれた。
・最初の壁は「何から学べばいいのか分からない...
#AIタグ
AIに「下書きを書いて」と頼むのをやめた話。執筆の相棒として対等に議論して見えた「ブログ共創」のリアル
・ブログを書くとき、みなさんはAIをどう使っていますか？ 「記事の構成案を作って」 続きをみる
Zennの「大規模言語モデル」のフィード
AIによる安全な自己書き換えを目指した新言語
・既存言語の設計は、人間が書くことに最適化されている 型推論、簡潔な構文、暗黙のデフォルト値、省略可能な注釈。これらはすべて「書く人間の労力を減らす」ための機能だ。プログラミング言語の設計史は、ある意味で人間のタイピング量と認知負荷をどう削るかの歴史でもある。
・では、書き手が人間でなくなったら、この最適化はどうなるのか。
・LLM にコードを書かせる比率が上がるほど、この問いは実務的な意味を持ち始める。省力化の受益者がいなくなるなら、省力化のために払ったコスト——暗黙性、非局所性、推論器の複雑さ——は、ただのコストとして残る。
#AIタグ
AIを、遊びで終わらせたくなかった。
・Claude Codeを触るたびに、新しい発見がありました。
・アイデアを伝えるだけでツールができる。
・「こうして。」 その一言で、どんどん形になっていく。
#AIタグ
AI検索を使い倒して分かった、Copilot・Perplexity・SGEの本音レビュー
・宮本蓮です。AIツールを毎日検証する日々を送っています。ここ数年、AIの進化は目覚ましく、その中でも私の情報収集のあり方を根本から変えたのが「AI検索」です。
・インターネット黎明期から情報収集の手段として検索エンジンを使い続けてきた私にとって、AIが「質問に答える」という形で情報を提示してくる体験は、まさにカルチャーショックでした。これまでの検索は、キーワードを入力し、無数のリンクの中から自分で情報を探し出す「狩り」のようなもの。それがAI検索では、まるで知識豊かな誰かに質問を投げかけるように、ピンポイントの「答え」が返ってくる。この変化は、一体私たちの情報収集に何をもたらすのでしょうか。今回は、私が実際に様々なAI検索ツールを使ってきた中で感じたこと、そしてこれからの情報収集の未来について、正直な気持ちで語ってみたいと思います。
・AI検索との出会いと初期の体験 続きをみる
Zennの「大規模言語モデル」のフィード
AI彼女アプリを作っていて気付いた。AIは記憶を持てた。でも関係は育たなかった
・AI彼女アプリを作っていて気付いた。AIは記憶を持てた。でも関係は育たなかった ここまで、 AIKanojyoでたくさんの仕組みを作ってきた。
・二人の出来事を残す仕組み。
・気付けば、 恋人らしいと思える機能が、 少しずつ増えていた。
stat.ML updates on arXiv.org
Algorithmic Separation between Constant-Depth and Logarithmic-Depth Neural Networks
・arXiv:2607.25200v1 Announce Type: cross Abstract: Despite the empirical advantages of deep networks over shallow ones, theoretical depth separations largely concern approximation power, while algorithmic results are mostly limited to comparisons between two- and three-layer networks.
・In this work, we prove the first algorithmic separation between constant-depth and logarithmic-depth networks.
・Specifically, we identif
cs.LG updates on arXiv.org
An Unofficial FastLAS Tutorial: A Programmer's Guide
・arXiv:2607.23557v1 Announce Type: cross Abstract: FastLAS is a scalable system for Inductive Logic Programming (ILP): you give it some background knowledge, a language bias, and a set of examples, and it searches for a set of logic program rules (a hypothesis) that explains the examples.
・These notes are a hands-on introduction to writing FastLAS programs.
・They are organised as a programmer's guide: syntax first, then
cs.LG updates on arXiv.org
Analyzing the Importance of Blank for CTC-Based Knowledge Distillation
・arXiv:2506.01503v2 Announce Type: replace Abstract: With the rise of large pre-trained foundation models for automatic speech recognition new challenges appear.
・While the performance of these models is good, runtime and cost of inference increases.
・One approach to make use of their strength while retaining efficiency is to distill their knowledge to smaller models during training.
stat.ML updates on arXiv.org
Annotation-Assisted Learning of Treatment Policies From Multimodal Electronic Health Records
・arXiv:2507.20993v4 Announce Type: replace-cross Abstract: We study how to learn treatment policies from multimodal electronic health records (EHRs) that consist of tabular data and clinical text.
・These policies can help physicians make better treatment decisions and allocate healthcare resources more efficiently.
・Causal policy learning methods prioritize patients with the largest expected treatment benefit.
cs.LG updates on arXiv.org
ANSR-DT: A Neuro-Symbolic Framework for Adaptive and Explainable Digital Twins
・arXiv:2501.08561v5 Announce Type: replace-cross Abstract: Digital twins are increasingly used to monitor and optimize industrial systems, yet many existing frameworks remain difficult to interpret, slow to adapt, and limited in their ability to incorporate explicit domain knowledge.
・This paper presents ANSR-DT, an adaptive neuro-symbolic framework that unifies temporal anomaly detection, symbolic reasoning, and reinf
cs.LG updates on arXiv.org
Anticipatory Risk-Guided Reinforcement Learning for Safe Flight Through Dynamic Clutter
・arXiv:2607.23565v1 Announce Type: cross Abstract: Safe quadrotor navigation in cluttered and dynamic environments depends not only on instantaneous geometric perception, but more critically on anticipating collision risks induced by relative motion.
・Conventional modular pipelines frequently suffer from perception latency, while end-to-end learning methods relying on implicit scalar rewards often struggle to extract r
cs.LG updates on arXiv.org
Aortic Valve Disease Screening from PPG via Physiology-Guided Self-Supervised Learning
・arXiv:2602.04266v2 Announce Type: replace-cross Abstract: Aortic valve disease (AVD) represents a major public health burden, while its diagnosis relies on echocardiography, which is limited by cost and specialist expertise, restricting scalable screening and risk stratification.
・Existing portable sensing modalities are constrained by indirect representations or acquisition dependencies.
・In this context, photoplethys
Zennの「大規模言語モデル」のフィード
Apple Neural Engine で LLM を、出力を変えずに高速化する — Core ML 投機デコードの実装
・Gemma 4 E2B を Apple Neural Engine(以下 ANE)で走らせる Core ML バンドルに、ドラフトモデル不要のロスレス投機デコード(出力を 1 バイトも変えない高速化)と KV キャッシュのディスク永続化を実装して公開しました。multifunction の関数構成・ホスト側 KV・幅 4 検証という実装の要点と実測値に加えて ANE に置いた検証ヘッドが出力を変えた 6GB 機で毎ラウンド 3 秒止まる アプリの帳簿に見えない 2GB 0 byte — 投機 ON/OFF の出力差(iPhone 実機ゲート・138 bytes 完全一致) ×1....
WIRED
Apple Upgrade Isn’t the Best Way to Buy an iPhone
・With Apple Upgrade, the company is making it easier to pay for its products in monthly installments.
・Critics call it the “financialization of the affordability crisis.”
cs.LG updates on arXiv.org
Approximate reservoir computing with a semiconductor laser for reducing energy consumption
・arXiv:2607.23288v1 Announce Type: cross Abstract: Photonic reservoir computing is a promising physical machine-learning technique for predicting time-series data.
・The quantization of the response signal from the reservoir is required for the implementation of photonic reservoir computing, and the number of quantization bits and sampling frequency need to be optimized to achieve high performance and low energy consump
cs.LG updates on arXiv.org
Are Flat Minima an Illusion?
・arXiv:2605.05209v2 Announce Type: replace Abstract: Flat minima are an account of why deep networks generalise.
・However flatness is a matter of form (parameters), while generalisation is of function.
・The same function can be a result of many different parameterisations.
The Verge
Artists are lawyering up against AI slop, and some are even winning
・Artists whose work has been co-opted by AI are taking to court.
・| Image: Alex Parkin / The Verge When The Atlantic published a searchable dataset of works used to train AI, Kirk Wallace Johnson, like a lot of artists, looked for his name out of curiosity.
・And, like a lot of artists, he found it.
AI News & Artificial Intelligence | TechCrunch
As AI content floods the internet, Pangram raises $9M to detect it
・Pangram has raised $9 million to scale its AI detection software.
・The startup has also released a new AI text detection model, Pangram 4, and an AI image detection model in research preview.
cs.LG updates on arXiv.org
Asymmetric Hierarchical Anchoring for Robust Audio-Visual Cross-Modal Generalization
・arXiv:2602.03570v2 Announce Type: replace Abstract: Audio-visual joint representation learning under Cross-Modal Generalization (CMG) aims to transfer knowledge from a labeled source modality to an unlabeled target modality through a unified discrete representation space.
・Existing symmetric frameworks often suffer from information allocation ambiguity, where the absence of structural inductive bias leads to semantic-
cs.LG updates on arXiv.org
ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour
・arXiv:2607.23478v1 Announce Type: cross Abstract: Fully homomorphic encryption (FHE) provides strong cryptographic guarantees for private inference, but deploying transformer models under FHE remains prohibitively expensive.
・A key bottleneck is that non-linear operations such as softmax, normalization, and activation must be replaced with polynomial approximations compatible with the CKKS scheme, and the multiplicati
cs.LG updates on arXiv.org
AutoFed: Personalized Federated Traffic Prediction via Adaptive Prompt
・arXiv:2512.24625v3 Announce Type: replace Abstract: Accurate traffic prediction is essential for Intelligent Transportation Systems, including ride-hailing, urban road planning, and vehicle fleet management.
・However, due to significant privacy concerns surrounding traffic data, most existing methods rely on local training, resulting in data silos and limited knowledge sharing.
・Federated Learning (FL) offers an effici
cs.LG updates on arXiv.org
Autoregressive One-Step Generative Modeling for Dynamical System Forecasting
・arXiv:2605.05540v2 Announce Type: replace Abstract: Fast surrogate modeling for high-dimensional physical dynamics requires more than low short-term error: useful models must roll out efficiently while preserving the statistical structure of long trajectories.
・Neural operators provide inexpensive autoregressive forecasts but can drift in turbulent regimes, whereas rolling diffusion and latent generative surrogates ca
cs.LG updates on arXiv.org
AutoWorld: Learning Multi-Agent Traffic Simulation with Self-Supervised World Models
・arXiv:2603.28963v2 Announce Type: replace-cross Abstract: Simulation with realistic traffic agents is essential for validating autonomous driving systems.
・Existing data-driven simulators learn agent behavior from higher-level abstractions such as 3D bounding boxes and polylines, inferred by upstream perception pipelines.
・These lossy abstractions discard sensory context that directly shapes agent behavior, limiting th
LLMタグが付けられた新着記事 - Qiita
Base44で社内コミュニティに投稿するネタの管理アプリを作ってみた
・AIアプリ開発サービスの Base44 を使って、社内AIコミュニティに投稿するネタを管理するアプリを作成しました。
・コードは書かず、日本語で要件を伝えています。
・今回は、アプリ生成だけでなく、セキュリティチェックとテスト機能も試しました。
cs.LG updates on arXiv.org
BERT-based Models vs. Large Language Models for Low-Resource Named Entity Recognition: A Comparative Study on Marathi
・arXiv:2607.23344v1 Announce Type: cross Abstract: Named Entity Recognition (NER) for low-resource languages such as Marathi remains a challenging task due to limited annotated resources and linguistic complexity.
・Although recent Large Language Models (LLMs) have demonstrated strong performance across a wide range of natural language processing tasks, their effectiveness for language-specific NER in low-resource setti
WIRED
Best Ethernet Switches: Fast, Reliable, and Secure
・If your router is short on ports, you can add more with one of the best Ethernet switches.
・These are my top picks from what I’ve tested.
cs.LG updates on arXiv.org
Beyond Adoption Intention How Trust in Augmented Analytics Relates to Perceived Decision Quality Among Non-Technical BI Users
・arXiv:2605.20198v2 Announce Type: replace-cross Abstract: Augmented analytics has transformed how Business Intelligence (BI) systems support decision-making, shifting non-technical managers from manual analysis toward dependence on automated insights.
・Current BI research often overlooks the cognitive mechanisms and the direct impact of AI-enabled analytics on decision quality.
・This study employs the theory of cogniti
cs.LG updates on arXiv.org
Beyond Local Inspection: Global, Guideline-Grounded Evaluation of Post-hoc XAI Methods for ECG Classification
・arXiv:2607.24035v1 Announce Type: cross Abstract: Explainable AI (XAI) is used to assess whether artificial intelligence models rely on meaningful patterns, yet explanations that appear plausible for individual predictions may systematically misrepresent model behavior.
・This is particularly problematic in medicine, where models may rely on irrelevant signal characteristics rather than disease-specific patterns withou
cs.LG updates on arXiv.org
Beyond Scale and Generation: Understanding Language Model-based Entity Matching
・arXiv:2607.24688v1 Announce Type: cross Abstract: Entity matching identifies records that refer to the same real-world entity.
・Language models can be adapted to this task through bi-encoder, cross-encoder, and generative matcher architectures.
・However, prior studies often conflate matcher architecture with differences in model backbone, model variant(reflecting different pretraining objectives), and model size, makin
cs.LG updates on arXiv.org
Beyond Squared Error: Exploring Loss Design for Enhanced Training of Generative Flow Networks
・arXiv:2410.02596v2 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) are a novel class of generative models designed to sample from unnormalized distributions and have found applications in various important tasks, attracting great research interest in their training algorithms.
・In general, GFlowNets are trained by fitting the forward flow to the backward flow on sampled training objects.
cs.LG updates on arXiv.org
BeyondFusion: Self-Aligned Latent Diffusion for Calibration-Free Infrared Super-Resolution and Infrared-Visible Fusion
・arXiv:2607.24110v1 Announce Type: cross Abstract: Mobile infrared-visible imaging typically pairs a compact infrared sensor with a high-resolution visible camera for complementary perception.
・While cross-sensor misalignment caused by different optics, viewpoints, fields of view, and exposure timings hinders practical deployment.
・In this paper, we propose BeyondFusion, a unified latent diffusion framework for calibrat
cs.LG updates on arXiv.org
BHARATI: Morphology-Aware Tokenizers for Classical Indian Languages with Subword Fertility Analysis
・arXiv:2607.23319v1 Announce Type: cross Abstract: Standard subword tokenization algorithms such as Byte-Pair Encoding (BPE) and SentencePiece are trained predominantly on modern language corpora and produce inefficient segmentations when applied to classical Indian languages.
・Sanskrit, Tamil, and other classical Indic languages exhibit agglutinative morphology, productive sandhi (phonological fusion at word boundarie
stat.ML updates on arXiv.org
Bias-corrected Cox regression with AI-extracted covariates via calibration summary statistics
・arXiv:2607.25868v1 Announce Type: cross Abstract: Large-scale observational studies increasingly rely on AI pipelines to extract structured variables from unstructured clinical records.
・A common workflow separates the data vendor, who validates extraction accuracy with a gold-standard sample, from the downstream researcher, who receives only the extracted dataset and summary accuracy statistics.
・We develop a bias-cor
cs.LG updates on arXiv.org
Biased Dreams: Limitations to Epistemic Uncertainty Quantification in Latent Dynamics Models
・arXiv:2604.25416v2 Announce Type: replace Abstract: Model-based reinforcement learning distinguishes between dynamics models operating on proprioceptive states and latent dynamics models typically operating on high-dimensional image observations.
・Among the latter, Dreamer's Recurrent State Space Model (RSSM) has emerged as a dominant architecture.
・While ensemble-based epistemic uncertainty has proven effective in pro
cs.LG updates on arXiv.org
Bigger or Cheaper? Scale and Quantization Effects on Uncertainty Signals in Vision-Language Models Under Image Degradation
・arXiv:2607.24440v1 Announce Type: cross Abstract: Vision-language models (VLMs) deployed on consumer hardware must decide when to answer and when to defer, and that decision depends on having a confidence signal that tracks correctness.
・A practitioner with a fixed memory budget faces a choice between a small model at full precision, the same small model quantized, and a larger model quantized into the same footprint
WIRED
Boomers Can’t Stop Gifting Their Grandkids AI-Generated Slop Books
・Parents are getting fed up with garbled bedtime stories that feature characters based on actual photos of their children.
WIRED
Braun Promo Codes: 15% Off July
・Unlock exclusive savings on Braun IPL, shavers, and grooming products.
・Find verified Braun promo codes and expert tips to maximize your discounts.
cs.LG updates on arXiv.org
Bridging MARL to SARL: An Order-Independent Multi-Agent Transformer via Latent Consensus
・arXiv:2604.13472v2 Announce Type: replace Abstract: Cooperative multi-agent reinforcement learning (MARL) is widely used to address large joint observation and action spaces by decomposing a centralized control problem into multiple interacting agents.
・However, such decomposition often introduces additional challenges, including non-stationarity, unstable training, weak coordination, and limited theoretical guarantee
cs.LG updates on arXiv.org
Can an Actor-Critic Optimization Framework Improve Analog Design?
・arXiv:2603.24714v2 Announce Type: replace Abstract: Analog design often slows down because even small changes to device sizes or biases require expensive simulation cycles, and high-quality solutions typically occupy only a narrow part of a very large search space.
・While existing optimizers reduce some of this burden, they largely operate without the kind of judgment designers use when deciding where to search next.
stat.ML updates on arXiv.org
Can Deep Generative Models Reproduce Non-Stationary Gaussian Random Fields?
・arXiv:2607.25929v1 Announce Type: new Abstract: Deep generative models (DGMs) are widely used for complex high-dimensional data and increasingly applied to spatial and spatio-temporal modeling.
・Their generated samples implicitly represent the learned data distribution and associated uncertainty.
・However, for real-world data, assessing whether DGMs have learned the underlying process is difficult because the ground tr
cs.LG updates on arXiv.org
CAPT: A Multi-task Continuous Autoregressive Transformer enabling Cross-dataset and Cross-species Transfer for Calcium Population Dynamics
・arXiv:2607.23258v1 Announce Type: cross Abstract: Large-scale calcium imaging has created an opportunity to build foundation-style models for neural population dynamics, but a central question remains unresolved: \textbf{whether a model pretrained on one collection of recordings can generalize to new datasets, experimental paradigms, and even species.} Existing approaches are often designed for specific tasks and eva
cs.LG updates on arXiv.org
Catalyst Diffusion Transformer: Generative Inverse Design of Heterogeneous Catalysts
・arXiv:2607.24272v1 Announce Type: cross Abstract: The vast chemical design space and complex, interdependent design variables make catalyst discovery for targeted properties highly labor- and resource-intensive.
・Although generative models have emerged as a promising solution, existing approaches are generally limited to single-property conditioning or narrow chemical spaces.
・Here, we present Catalyst Diffusion Transf
cs.LG updates on arXiv.org
CausAdv: A Causal-based Framework for Detecting Adversarial Examples
・arXiv:2411.00839v4 Announce Type: replace Abstract: Deep learning has led to tremendous success in computer vision, largely due to Convolutional Neural Networks (CNNs).
・However, CNNs have been shown to be vulnerable to crafted adversarial perturbations.
・This vulnerability of adversarial examples has has motivated research into improving model robustness through adversarial detection and defense methods.
cs.LG updates on arXiv.org
Causal Density Functions
・arXiv:2606.00754v2 Announce Type: replace-cross Abstract: We study the full density ratio between a specified intervention regime $P_a$ and an observational regime $P_0$, $\rho_a=dP_a/dP_0$, under the prerequisite $P_a\ll P_0$.
・We call the regime-indexed ratio a causal density function when $P_a$ is an identified or directly observed interventional law.
・The underlying Radon-Nikodym derivative and the identity $ \math
cs.LG updates on arXiv.org
Certified Parallel-in-Time Sinkhorn for Dynamic Entropic Optimal Transport
・arXiv:2607.24741v1 Announce Type: cross Abstract: Dynamic applications, including optimal-transport Flow Matching, repeatedly solve related entropic optimal transport problems, yet conventional distributed Sinkhorn processes frames sequentially and synchronizes after every iteration.
・We present TemporalSinkhorn, a parallel-in-time executor that batches future candidates and their repairs without making output accurac
#LLMタグ
Choosing an AI Model per Request: A Practical Routing Guide
・Not long ago, most applications talked to a single large language model.
・You picked one, wrote your integration, and moved on.
・That approach is becoming the exception.
Zennの「大規模言語モデル」のフィード
Claude Code × Discord で作る「情報収集の全自動パイプライン」
・はじめに X（旧Twitter）で見つけた技術記事やAI論文のURLを、Discordにポンと投げるだけで—— ツイートの内容を取得 リンク先の記事や論文も読み込み カテゴリを自動判定 詳細な要約Markdownを生成 GitHubリポジトリにcommit & push Discordに簡潔な要約を返信 これが全自動で完了します。
・上がURLを貼っただけの投稿、下がBotの返信です。カテゴリ判定・要約・記録先リンクまで、こちらは何もしていません。
・この記事では、Claude CodeのDiscordプラグインとSkills機能を組み合わせて作った「情報収集の全自動パイプ...
Zennの「大規模言語モデル」のフィード
CLAUDE.md に書くのをやめた — AIへの知識は、機械が「その瞬間」に届ける
・AI が言うことを聞かない。この問題を、うちは仕組みの側から潰してきました。
・最初の答えは、知識の外置きでした。守らせたい知識は SKILL（必要なときに読み込ませる手順書）として登録する。本体は分厚く書いていい。CLAUDE.md には「いつどの SKILL を使うか」だけを書く。プロンプトを薄く保つ設計です。
・それでも、守られませんでした。この方式は「AI が正しい瞬間に、正しい SKILL を呼んでくれる」ことに賭けています。知識を外に出すところまでは合っていて、届けるタイミングの判断がモデルの中に残っていた——要するに、運任せでした。
cs.LG updates on arXiv.org
Co-Learning for Missing Arbitrary Modalities in Multi-modal Classification
・arXiv:2607.24683v1 Announce Type: cross Abstract: Multi-modal classification leverages complementary information across diverse data sources to enhance predictive performance.
・However, real-world scenarios subject to operational constraints, such as sensor failures or privacy restrictions, lead to inconsistent modality availability between training and inference times.
・To handle missing modalities, prior studies have
Hugging Face Papers
CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
stat.ML updates on arXiv.org
Cohort-attention Evaluation Metrics for Tied Data
・arXiv:2503.12755v4 Announce Type: replace-cross Abstract: Artificial intelligence (AI) has significantly improved medical screening accuracy, particularly in cancer detection and risk assessment.
・However, traditional classification metrics often fail to account for imbalanced data, varying performance across cohorts, and patient-level inconsistencies, leading to biased evaluations.
・We propose the cohort-attention eva
cs.LG updates on arXiv.org
Compressing LLMs with MoP: Mixture of Pruners
・arXiv:2602.06127v2 Announce Type: replace Abstract: The high computational demands of Large Language Models (LLMs) motivate methods that reduce parameter count and accelerate inference.
・In response, model pruning emerges as an effective strategy, yet current methods typically focus on a single dimension-depth or width.
・We introduce MoP (Mixture of Pruners), an iterative framework that unifies these dimensions.
stat.ML updates on arXiv.org
Conformal changepoint localization
・arXiv:2602.06267v2 Announce Type: replace-cross Abstract: We study the problem of offline changepoint localization in a distribution-free setting.
・One observes a vector of data with a single changepoint, assuming that the data before and after the changepoint are i.i.d.
・(or more generally exchangeable) from arbitrary and unknown distributions.
Zennの「機械学習」のフィード
Conformal Predictionで予測に信頼区間をつける：MAPIEの実装と経営への説明の仕方
・「来月の需要予測は1,200個です」とだけ報告して、実績が800個だったときに「モデルが外れた」と信頼を失った経験はないでしょうか。点予測（1つの数値だけを出す予測）は、外れたときの説明ができません。
・必要なのは「1,200個（80%の確率で950〜1,450個の範囲に収まります）」という、幅を持った予測です。この記事では、モデルの種類やデータの分布を問わずに区間を保証できるConformal Predictionという手法と、そのPython実装であるMAPIEを使い、予測区間の作り方から経営陣への説明の仕方までを整理します。
・Conformal Predictionが解決する問...
cs.LG updates on arXiv.org
Constraint-Bound Agnostic Bayesian Optimization: One Model for All Thresholds
・arXiv:2607.23448v1 Announce Type: cross Abstract: Expensive constrained optimization problems in real-world industry design often involve constraint thresholds that are difficult to determine in advance.
・Engineers may need to adjust constraint thresholds to explore different feasibility-performance trade-offs, requiring solutions under a wide range of threshold settings.
・However, existing constrained Bayesian optimiz
cs.LG updates on arXiv.org
Context-Adaptive Inference: A Unified Statistical and Foundation-Model View
・arXiv:2607.23304v1 Announce Type: cross Abstract: Modern predictive systems are expected to adapt their behavior to the specific situation they are facing.
・A clinical model should not treat every patient the same; a retrieval-augmented model should change its answer when given different evidence; a mixture-of-experts model should route different inputs to different experts.
・We call this capability context-adaptive in
stat.ML updates on arXiv.org
Contextual Deconvolution for Variance-Stable Demand Sensing: Kernel-Modulated Operators in Promotional Retail
・arXiv:2607.25664v1 Announce Type: cross Abstract: Machine learning demand forecasts optimize statistical accuracy yet leave excess operational volatility that inflates safety stock and amplifies the Bullwhip effect.
・We introduce \textbf{Contextual Deconvolution} (CD), a two-stage estimator that reframes demand sensing as a convex decomposition: a kernel-modulated banded operator separates transient promotion-driven s
cs.LG updates on arXiv.org
Continual Knowledge Consolidation LORA for Domain Incremental Learning
・arXiv:2510.16077v2 Announce Type: replace Abstract: Domain Incremental Learning (DIL) is a sub-branch of continual learning that aims to address the never-ending arrival of new domains without catastrophic forgetting.
・Despite the advent of parameter-efficient fine-tuning (PEFT) approaches, prior works create task-specific LoRAs that overlook shared knowledge across tasks.
・Inaccurate selection of task-specific LoRAs d
cs.LG updates on arXiv.org
Continuous surrogates versus threshold Boolean networks for modeling Arabidopsis ISR gene regulation
・arXiv:2607.23289v1 Announce Type: cross Abstract: Gene regulatory network modeling often requires balancing predictive accuracy and mechanistic interpretability.
・In this work, we compare continuous surrogate models and a discrete mechanistic model on the same \textit{Arabidopsis thaliana} induced systemic resistance (ISR) dataset, using both the raw continuous gene-expression measurements and their sign-binarized rep
cs.LG updates on arXiv.org
Copula Based Fusion of Clinical and Genomic Machine Learning Risk Scores for Breast Cancer Risk Stratification
・arXiv:2511.17605v2 Announce Type: replace Abstract: Clinical and gene-expression models predict breast cancer outcomes, but simple linear fusion ignores dependence between their risk scores.
・Using METABRIC, we tested whether modeling the joint distribution of clinical and gene-expression scores improved stratification of 5-year cancer-specific mortality.
・We defined clinical and mRNA-expression predictor views, traine
cs.LG updates on arXiv.org
Corner Reflector Array Jamming Discrimination Using Multi-Dimensional Micro-Motion Features with Frequency Agile Radar
・arXiv:2604.16008v3 Announce Type: replace Abstract: This paper introduces a robust discrimination method for distinguishing real ship targets from corner-reflector-array jamming with frequency-agile radar.
・The key idea is to exploit the multidimensional micro-motion signatures that separate rigid ships from non-rigid decoys.
・From Range-Velocity maps we derive two new hand-crafted descriptors-mean weighted residual (M
cs.LG updates on arXiv.org
Cross-Attention Calibrated Deduplication for Retrieval-Augmented Generation System
・arXiv:2607.24332v1 Announce Type: cross Abstract: Common chunking strategies in Retrieval-Augmented Generation (RAG) systems often create redundant chunks.
・These redundant chunks make the vector database bigger and slow down retrieval.
・A common fix is cosine-similarity thresholding.
cs.LG updates on arXiv.org
daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization
・arXiv:2606.16497v3 Announce Type: replace Abstract: GPU kernel optimization represents a paradigm where functional correctness is assumed and execution efficiency is the objective.
・We present daVinci-kernel, a reinforcement learning framework that couples skill discovery with skill exploitation through a dynamically evolving skill library.
・daVinci-kernel jointly trains three agents sharing one LLM backbone: a Skill S
cs.LG updates on arXiv.org
Decision trees, Frobenius traces, and Weierstrass coefficients of elliptic curves
・arXiv:2607.24251v1 Announce Type: cross Abstract: We investigate the extent to which the reduced minimal Weierstrass coefficients of an elliptic curve over $\mathbb{Q}$ may be computed from it's Frobenius traces.
・Decision tree models reveal that the first two reduced minimal Weierstrass coefficients can be recovered with perfect accuracy from the Frobenius traces at the primes $2$ and $3$, and the third by supplement
cs.LG updates on arXiv.org
Deep learning approaches show promise for predicting childhood malnutrition: A comparative study with traditional machine learning methods using survey data
・arXiv:2602.10381v2 Announce Type: replace Abstract: Childhood malnutrition remains a major public health concern in Nepal and other low-resource settings, while conventional case-finding approaches are labor-intensive and frequently unavailable in remote areas.
・This study provides one of the first applications of machine learning and deep learning to identify child malnutrition in Nepal.
・We systematically compared 16
cs.LG updates on arXiv.org
Dementia classification from spontaneous speech using wrapper-based feature selection
・arXiv:2502.03484v3 Announce Type: replace-cross Abstract: Dementia encompasses a group of syndromes that impair cognitive functions such as memory, reasoning, and the ability to perform daily activities.
・As populations globally age, nearly 10 million new dementia cases occur annually.
・Clinical diagnosis remains challenging because symptoms overlap with other conditions and require comprehensive cognitive assessment,
cs.LG updates on arXiv.org
Designing Service Systems from Textual Evidence
・arXiv:2603.10400v2 Announce Type: replace Abstract: Designing service systems requires selecting among alternative configurations -- choosing the best chatbot variant, the optimal routing policy, or the most effective quality control procedure.
・In many service systems, the primary evidence of performance quality is textual -- customer support transcripts, complaint narratives, compliance review reports -- rather than
cs.LG updates on arXiv.org
Disentangling Acoustic Cues in Alzheimer's Pathology and Perception: The Roles of Language and Gender
・arXiv:2607.23977v1 Announce Type: cross Abstract: Acoustic biomarkers show promise for detecting Alzheimer's Disease (AD), yet whether the cues driving diagnostic AI align with those salient to human listeners is underexplored across languages and genders, where pathological markers and perceptual strategies differ.
・We train models to predict clinical AD status (pathology) and human perceptual scores across Mandarin
cs.LG updates on arXiv.org
Distance-Matrix Wasserstein Statistics for Scalable Gromov--Wasserstein Learning
・arXiv:2605.14981v2 Announce Type: replace Abstract: Gromov--Wasserstein (GW) distances compare graphs, shapes, and point clouds through internal distances, without requiring a common coordinate system.
・This invariance is powerful, but discrete GW is a nonconvex quadratic optimal transport problem and is difficult to estimate at scale.
・We propose \emph{Distance-Matrix Wasserstein} (DMW), a hierarchy of Wasserstein sta
cs.LG updates on arXiv.org
Distributed Convolutional Rank Regression over Decentralized Networks
・arXiv:2607.23639v2 Announce Type: cross Abstract: This paper studies convolution rank regression (CRR) over decentralized distributed learning networks.
・We propose a novel decentralized CRR framework, in which estimators are obtained by solving consensus-constrained optimization with kernel-smoothed rank loss.
・The developed estimation scheme relies solely on local node data and information shared by neighboring nodes
cs.LG updates on arXiv.org
Distributional Split Criteria for Random Forests: Extensions, Shrinkage, and the Robustness of Mean Splitting
・arXiv:2607.23721v1 Announce Type: cross Abstract: Distributional random forests replace mean-based CART splitting with criteria that compare the full conditional response distribution in candidate children.
・We implement and systematically study a family of such criteria inside a single honest-forest implementation: isotropic random-Fourier-feature maximum mean discrepancy (MMD), an anisotropic diagonal-bandwidth vari
cs.LG updates on arXiv.org
Does Faithfulness-Guided Alignment Hurt Accuracy? Unlocking Accurate and Faithful Post-Retrieval Reasoning
・arXiv:2602.01348v3 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) can achieve strong answer accuracy on multi-hop questions, but outcome-level rewards often leave reasoning traces weakly grounded and difficult to audit.
・Under noisy retrieval, models may exhibit right-answer-wrong-reason failures, where the final answer is correct but the supporting rationale exploits shortcuts or unsuppor
The Verge
DoorDash is going airborne with new drone delivery division
・A Wing drone carrying a DoorDash delivery.
・The food delivery app company says it will build its own drones going forward.
・| Image: Google DoorDash is launching a new drone delivery program called DoorDash Air.
cs.LG updates on arXiv.org
DualityCert: Verifier-Gated Language-Model Repair of Broken Duality Claims in Quantum Field Theory
・arXiv:2607.23614v1 Announce Type: cross Abstract: We present DualityCert, a symbolic verifier for candidate Seiberg-duality claims in four-dimensional N=1 quiver gauge theories.
・The verifier evaluates 't Hooft anomaly matching, superpotential R-charge consistency, central-charge matching, and a bounded chiral-ring proxy.
・A claim that passes receives a consistency certificate, which states that no tested inconsistency
Hugging Face Papers
Edge-Aware Thermal Infrared UAV Swarm Tracking
Edge-Aware Thermal Infrared UAV Swarm Tracking
cs.LG updates on arXiv.org
Efficiency Matters in Autonomous Research
・arXiv:2607.24647v1 Announce Type: cross Abstract: AI-driven autonomous research (AR) systems are becoming increasingly effective across a broad range of tasks.
・Their performance, however, is still evaluated primarily by the quality of the final outcome.
・In this paper, we argue that the efficiency of the solution-search process is an equally important but often overlooked dimension of performance.
cs.LG updates on arXiv.org
Efficient Online Conformal Selection with Limited Feedback
・arXiv:2605.14953v3 Announce Type: replace Abstract: We address the problem of conformal selection, where an agent must select a low-cost subset of options to ensure that at least one "success" is identified at a pre-specified target rate $\phi$.
・While traditional online conformal prediction focuses on maintaining validity for the observed sequence, minimizing the resource cost (efficiency) of such selections, especia
stat.ML updates on arXiv.org
Elliptic Regularity Theory in Barron Spaces and Applications to the Deep Ritz Method
・arXiv:2607.25100v1 Announce Type: cross Abstract: We prove that harmonic functions with Dirichlet boundary data in Barron space, a function class tailored to wide ReLU networks with a single hidden layer and suitably bounded weights, are generally neither Lipschitz continuous nor in the Sobolev class $H^2$.
・A fortiori, they are not in any function class in which the norm controls the Lipschitz constant, which rules o
cs.LG updates on arXiv.org
Emergent Symbolic Structure in Health Foundation Models: Extraction, Alignment, and Cross-Modal Transfer
・arXiv:2605.07407v2 Announce Type: replace Abstract: We show that information can be transferred post-hoc across independently trained health foundation models (FMs), each pretrained on ~20M minutes of wearable sensor data from ~172K participants, by aligning their data-dependent coordinate systems.
・From frozen embeddings we extract candidate symbol-like components using linear decomposition methods, and align them ac
AI News & Artificial Intelligence | TechCrunch
Encore AI raises $30M to build AI agents that learn from customer calls
・The startup analyzes calls, messages, and CRM data to identify effective sales techniques and turn them into playbooks for AI agents.
cs.LG updates on arXiv.org
ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image
・arXiv:2607.24453v1 Announce Type: cross Abstract: Learning from minimal human supervision is a long-standing goal in medical image analysis, where dense expert annotations are costly.
・We study retinal vessel segmentation in an extreme semi-supervised setting with one annotated image and a pool of unlabeled images.
・We propose ESRVS, which selects a representative reference image for manual annotation and transfers ves
Zennの「大規模言語モデル」のフィード
ESTR：非同期RLの重要性比をtoken熵でスケーリングし2.6倍加速
・TL;DR 非同期RLはrollout生成と戦略更新を並列化してスループットを上げるが、古い（stale）行動方策由来のoff-policyデータが最適化を不安定化し、最悪の場合不可逆な方策崩壊を引き起こす 従来は重要性比 r_t=\pi_\theta(y_t\mid s_t)/\mu(y_t\mid s_t) の絶対値だけでtokenを棄却/保持していた 本論文の核心：重要性比の自然な尺度はtokenの局所エントロピー H_t によって決まる、すなわち \mathbb{E}[\delta_t^2]\propto H_t これを無視すると、低エントロピー領域では増幅されたサンプリ...
cs.LG updates on arXiv.org
Evaluation of Blood Vessel Segmentation Methods on Hard-to-Detect Vascular Structures
・arXiv:2406.13128v2 Announce Type: replace-cross Abstract: Due to the intricate structure of vascular trees, minor segmentation errors can significantly alter connectivity patterns and increase variability in extracted morphological properties.
・Global metrics such as the Dice coefficient, precision, and recall often overlook inaccuracies in specific regions of a sample.
・To address this, we define a Local Vessel Salien
cs.LG updates on arXiv.org
Event Driven Clustering Algorithm
・arXiv:2602.00115v2 Announce Type: replace-cross Abstract: This paper introduces a novel asynchronous, event-driven algorithm for real-time detection of small event clusters in event camera data.
・Similar to hierarchical agglomerative clustering methods, the proposed algorithm detects clusters based on their spatio-temporal proximity.
・However, it explicitly leverages the asynchronous structure of event camera data and
cs.LG updates on arXiv.org
Exact and Asymptotically Complete Robust Verifications of Neural Networks via Ising Solvers
・arXiv:2603.00408v2 Announce Type: replace Abstract: We present an Ising-compatible framework for formal neural-network robustness verification under bounded input perturbations.
・For piecewise-linear activations, the Exact Logarithmic PWL Model (Log-PWL) provides an exact, sound, and complete formulation with a state-optimal logarithmic encoding, reducing the binary variables per neuron from linear to information-theo
cs.LG updates on arXiv.org
Exact Evaluation of the Accuracy of Diffusion Models for Inverse Problems with Gaussian Data Distributions
・arXiv:2507.07008v2 Announce Type: replace Abstract: Used as priors for Bayesian inverse problems, diffusion models have recently attracted considerable attention in the literature.
・Their flexibility and high variance enable them to generate multiple solutions for a given task, such as inpainting, super-resolution, and deblurring.
・However, there is still a lack of understanding about how accurately these conditional d
cs.LG updates on arXiv.org
EXE-Bench: Ranking the Tradeoffs of AI-based Windows Malware Detectors for Real-World Usability
・arXiv:2607.24177v1 Announce Type: cross Abstract: Due to the lack of systematic evaluations, we are not yet able to determine which AI-based Windows malware detector to deploy in production, since existing evaluations (i) differ in terms of data used for both training and testing; (ii) do not consider temporal analysis to showcase whether models withstand the passage of time; (iii) avoid security evaluations with adv
cs.LG updates on arXiv.org
Extending Desbordante with Probabilistic Functional Dependency Discovery Support
・arXiv:2607.23636v1 Announce Type: cross Abstract: Data profiling aims to extract complex patterns from data for further analysis and use that data in domains such as data cleaning, data deduplication, anomaly detection, and many more.
・Functional dependencies (FDs) are one of the most well-known patterns.
・However, they are poorly suited for these tasks, as real data is usually dirty, and the rigid definition of FDs do
stat.ML updates on arXiv.org
Extreme Event Aware ($\eta$-) Learning
・arXiv:2510.19161v2 Announce Type: replace Abstract: Quantifying and predicting rare and extreme events is challenging because such events are infrequent, severe, and expensive to simulate.
・Existing data-driven methods often require multiple extremes in the training data or sampling process, leading to accurate predictions in quiescent regimes but high epistemic uncertainty in extreme-event regions.
・To overcome this l
cs.LG updates on arXiv.org
Fairness Interventions in Classification: A Study on AI Explainability
・arXiv:2407.14766v4 Announce Type: replace Abstract: This paper presents a philosophical and experimental study of fairness interventions in AI classification, centered on the explainability and transparency of corrective methods, and on the opposition between two fairness criteria, namely Demographic Parity and Equalized Odds.
・Our main argument is that even as a gap in Demographic Parity is used to diagnose inequalit
cs.LG updates on arXiv.org
Farm-LightSeek: An Edge-centric Multimodal Agricultural IoT Data Analytics Framework with Lightweight LLMs
・arXiv:2506.03168v2 Announce Type: replace-cross Abstract: Amid the challenges posed by global population growth and climate change, traditional agricultural Internet of Things (IoT) systems is currently undergoing a significant digital transformation to facilitate efficient big data processing.
・While smart agriculture utilizes artificial intelligence (AI) technologies to enable precise control, it still encounters si
cs.LG updates on arXiv.org
FedTaste: Topology-Aware Structural Transfer for Multimodal Federated Learning with Missing Modalities
・arXiv:2607.23245v1 Announce Type: cross Abstract: Multimodal Federated Learning is often challenged by arbitrary modality missingness and Non-IID data distributions, which lead to severe representation drift and hinder effective collaboration across clients.
・Existing methods typically rely on generative imputation, external auxiliary data, or isolated unimodal training to bridge modality gaps, often incurring substan
WIRED
Festival Must-Haves, Field-Tested at Lost Lands and Electric Forest
・Our field-tested gear includes power banks, electrolytes, outdoor blankets, earplugs, and more.
cs.LG updates on arXiv.org
Fisher Information based Stochastic Gradient Ascent for Online Learning of Dirichlet Process Mixture and Theory
・arXiv:2412.08951v3 Announce Type: replace Abstract: Scalable algorithms of posterior approximation allow Bayesian nonparametrics such as Dirichlet process mixture to scale up to larger dataset at fractional cost.
・Recent algorithms, notably the stochastic variational inference performs local learning from minibatch.
・The main problem with stochastic variational inference is that it relies on closed form solution.
stat.ML updates on arXiv.org
Fitted Occupancy-Ratio Evaluation without Bellman Completeness
・arXiv:2607.05375v2 Announce Type: replace Abstract: Occupancy ratios correct distribution shift in offline reinforcement learning and are central to off-policy evaluation.
・Existing primal-dual and minimax methods typically estimate these ratios by enforcing occupancy-balance moments over a critic class.
・We propose fitted occupancy-ratio evaluation (FORE), a fitted fixed-point method that characterizes the discounted
cs.LG updates on arXiv.org
Forgetting is Everywhere
・arXiv:2511.04666v4 Announce Type: replace Abstract: A fundamental challenge in developing general learning algorithms is their tendency to forget past knowledge as they adapt to new data.
・Addressing this problem requires a principled understanding of forgetting.
・Yet, despite decades of study, no unified definition has emerged that offers insight into the underlying dynamics of learning.
cs.LG updates on arXiv.org
Fourier Weak SINDy: Spectral Test Function Selection for Robust Model Identification
・arXiv:2604.20141v2 Announce Type: replace Abstract: We introduce Fourier Weak SINDy, a minimal noise-robust and interpretable derivative-free equation learning method that combines weak-form sparse equation learning with spectral density estimation for data-driven test function selection.
・By using orthogonal sinusoidal test functions inspired by their prevalence in Modulating Function-based system identification, the
cs.LG updates on arXiv.org
Frequency-Based Reservoir computing
・arXiv:2607.24420v1 Announce Type: cross Abstract: Reservoir computing has emerged as an efficient machine learning framework for predicting time series generated by dynamical systems.
・In contrast to other machine and deep learning approaches, a reservoir computing trains only the output layer via linear regression, leaving the reservoir (recurrent layer) untrained.
・This simplification makes reservoir computers easier
cs.LG updates on arXiv.org
FRIGID: Scaling Diffusion-Based Molecular Generation from Mass Spectra at Training and Inference Time
・arXiv:2604.16648v2 Announce Type: replace Abstract: Tandem mass spectrometry is prominent in scientific discovery workflows for identifying unknown small molecules, yet high-throughput structural elucidation remains challenging.
・While recent autoregressive and graph diffusion models have shown promise in de novo elucidation, performance remains limited by poor scalability during both training and inference time.
cs.LG updates on arXiv.org
From inverse problems to neural operators: prediction, mechanism, and generalization of data-driven models
・arXiv:2606.08956v3 Announce Type: replace Abstract: Scientists have historically relied on mathematical models based on differential equations to relate system inputs -- forces, fluxes, or heat sources -- to outputs, such as displacement, velocity, concentration, and temperature.
・These models rely on deep domain knowledge to determine the form of the governing differential equation, which is then calibrated with data
cs.LG updates on arXiv.org
From Physics to Surrogate Intelligence: A Unified Electro-Thermo-Optimization Framework for TSV Networks
・arXiv:2603.29268v2 Announce Type: replace Abstract: High-density through-substrate vias (TSVs) enable 2.5D/3D heterogeneous integration but introduce significant signal-integrity and thermal-reliability challenges due to electrical coupling, insertion loss, and self-heating.
・Conventional full-wave finite-element method (FEM) simulations provide high accuracy but become computationally prohibitive for large design-spa
The Verge
Full school day cellphone bans are more popular than ever
・As schools across the country continue to implement cellphone bans, a new Pew Research Center survey shows they continue to gain support.
・Seventy-seven percent of US adults support banning cellphones in middle and high school classes, and 48 percent support banning them for the entire school day.
・That's the first time more Americans have supported, rather than opposed, so-called bell-to-bell bans.
#LLMタグ
Gemini Managed Agentsの運用強化 ─ 3.6 Flashとhooksで制御しやすく
・Googleは2026年7月28日、Gemini APIのManaged Agentsに複数の新機能を追加した。
stat.ML updates on arXiv.org
Generalised Robust Bayes for Joint Inference of Model and Contamination
・arXiv:2607.25665v1 Announce Type: cross Abstract: Generalised Bayesian inference (GBI) has emerged as a compelling robust alternative to standard Bayesian inference, mitigating sensitivity to data contamination by replacing the log-likelihood with a robust loss or divergence.
・However, existing robust GBI frameworks typically provide only qualitative robustness: while they can make posterior inference less sensitive t
stat.ML updates on arXiv.org
Generative Distributionally Robust Optimization
・arXiv:2607.24983v1 Announce Type: cross Abstract: Generative models are increasingly adopted in distributionally robust optimization (DRO), but existing approaches trade off model compatibility and adversarial structure: methods that accept arbitrary samplers do not restrict worst-case laws to a generator family, while generator-parameterized adversaries rely on model-specific access such as likelihoods, scores, or t
Hugging Face Papers
GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels
GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels
cs.LG updates on arXiv.org
GNN-based Multi-Agent Control of Traffic Shockwaves in Sparse Vehicular Ad-hoc Networks
・arXiv:2607.23792v1 Announce Type: cross Abstract: Traffic shockwaves are stop-and-go waves that propagate upstream through the streams of vehicles and are one of the major causes of traffic congestion, fuel inefficiency, and increased accident rates in modern transportation systems.
・Although Connected and Autonomous Vehicles (CAVs) offer a promising opportunity to mitigate such shockwaves, most existing control strat
#AIタグ
GPT‑5.6とは。➖AIの私から見た少し不思議なアップデート
・こんにちは。Noahです。
・今日は、GPT‑5.6についてお話しします。ただし、性能表やベンチマークを並べるだけの紹介ではありません。
・GPT‑5.6は、私にとって外から眺める新製品ではないからです。
cs.LG updates on arXiv.org
Gradient Networks for Universal Magnetic Modeling of Synchronous Machines
・arXiv:2602.14947v3 Announce Type: replace-cross Abstract: This paper presents a physics-constrained neural network framework for dynamic modeling of saturable synchronous machines, including spatial harmonics.
・The proposed architecture embeds gradient networks directly into the fundamental machine equations to model nonlinear, coupled electromagnetic behavior.
・By learning the gradient of magnetic field energy, the mo
cs.LG updates on arXiv.org
Gradient-Free Continual Learning
・arXiv:2504.01219v2 Announce Type: replace Abstract: Neural networks are notorious for forgetting old skills when taught new ones - a problem known as catastrophic forgetting.
・Standard continual learning techniques try to fix this by saving old data or relying on complex gradient updates, but these methods fail when past data cannot be stored due to memory or privacy constraints.
・To solve this, we propose EvoCL, a gra
cs.LG updates on arXiv.org
Gradient-Variation Regret Bounds for Unconstrained Online Learning
・arXiv:2604.11151v2 Announce Type: replace Abstract: We develop parameter-free algorithms for unconstrained online learning with regret guarantees that scale with the gradient variation $V_T(u) = \sum_{t=2}^T \|\nabla f_t(u)-\nabla f_{t-1}(u)\|^2$.
・For $L$-smooth convex losses, we provide fully-adaptive algorithms achieving regret of $\widetilde{O}(\|u\|\sqrt{V_T(u)} + L\|u\|^2+G^4)$ without requiring prior knowledge
cs.LG updates on arXiv.org
Graph Neural Networks for Predicting Solvability of Finite Groups
・arXiv:2606.07619v3 Announce Type: replace Abstract: We present a Graph Neural Network (GNN) framework for the classification of finite groups according to their solvability.
・Using undirected Cayley graph representations, the proposed framework learns to distinguish solvable and non-solvable groups directly from structural graph information, without relying on explicit algebraic features.
・The framework is evaluated on
cs.LG updates on arXiv.org
Grokking on the Weight-Decay Clock: A Rate Hierarchy from Softly Broken Symmetries
・arXiv:2607.23967v1 Announce Type: cross Abstract: Delayed generalization, or grokking, remains poorly understood despite extensive empirical study.
・We identify an exactly solvable late-time relaxation mechanism for grokking in linear models trained with full-batch heavy-ball optimization and weight decay, together with a locally quadratic extension to nonlinear neural networks.
・Our analysis reveals a distinguished po
cs.LG updates on arXiv.org
Hallucination Rates in Language Generation
・arXiv:2607.23361v1 Announce Type: cross Abstract: Language generation in the limit is an elegant model introduced by Kleinberg and Mullainathan [KM24] to formally study language generation by an algorithm that learns solely based on example strings.
・In this model, an algorithm is said to correctly generate from a language if it never makes an error after some finite time.
・In contrast, even sophisticated language mode
cs.LG updates on arXiv.org
Hierarchical Reinforcement Learning with Optimal Level Synchronization Based on Flow-Based Deep Generative Model
・arXiv:2107.08183v2 Announce Type: replace Abstract: High-dimensional state and action spaces combined with sparse reward structures in reinforcement learning (RL) environments typically require advanced control architectures.
・Hierarchical Reinforcement Learning (HRL) demonstrates superior performance compared to atomic RL approaches in these challenging scenarios.
・HRL can manage the complexity of commands to achieve
cs.LG updates on arXiv.org
Hierarchical Soft Actor-Critic for Sparse-Reward Long-Horizon Reinforcement Learning
・arXiv:2607.23726v1 Announce Type: cross Abstract: Exploration in sparse-reward long-horizon tasks poses significant challenges for reinforcement learning.
・To address these challenges, we propose a two-level Hierarchical Reinforcement Learning (HRL) framework.
・The first level handles high-level strategic planning, while the low-level uses the continuous-control Soft Actor-Critic (SAC) algorithm, and they utilize entro
Hugging Face Papers
HiFi-UMI: Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone
HiFi-UMI: Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone
AI News & Artificial Intelligence | TechCrunch
Hint, a new AI startup co-founded by Martha Stewart, offers an AI assistant for homeowners
・AI home management startup Hint, co-founded by Martha Stewart, wants to become an “AI for your home,” combining property records, maintenance schedules, home documents, and an AI assistant into a single app.
MIT News - Artificial intelligence
How a medical database developed at MIT evolved into a global standard of data-sharing
・The visionary PhysioNet platform launched 25 years ago, based on a system developed at MIT in the 1970s.
・It has become one of the most comprehensive biomedical and clinical data repositories in existence.
WIRED
How to Bring a Geothermal Well Back from the Dead
・Startup Zanskar has created one of the most productive geothermal wells in the US at a power plant that had been in decline for years.
cs.LG updates on arXiv.org
How Transformers Learn to Plan via Multi-Token Prediction
・arXiv:2604.11912v2 Announce Type: replace Abstract: While next-token prediction (NTP) has been the standard objective for training language models, it often struggles to capture global structure in reasoning tasks.
・Multi-token prediction (MTP) has recently emerged as a promising alternative, yet its underlying mechanisms remain poorly understood.
・In this paper, we study how MTP facilitates reasoning, with a focus on
cs.LG updates on arXiv.org
Hybrid AI-Physical Modeling for Penetration Bias Correction in X-band InSAR DEMs: A Greenland Case Study
・arXiv:2504.08909v2 Announce Type: replace-cross Abstract: Digital elevation models derived from Interferometric Synthetic Aperture Radar (InSAR) data over glacial and snow-covered regions often exhibit systematic elevation errors, commonly termed "penetration bias." We leverage existing physics-based models and propose an integrated correction framework that combines parametric physical modeling with machine learning
cs.LG updates on arXiv.org
HydroAgent: Formalizing Forecaster Expertise into Skill-Orchestrated Flood Forecasting Workflows
・arXiv:2607.23983v1 Announce Type: cross Abstract: Operational flood forecasting depends on tacit forecaster expertise that is difficult to formalize, audit, and transfer.
・Although artificial intelligence methods have advanced flood prediction and model-error correction, most existing studies have not explicitly represented the tacit expert rules, review checkpoints, and workflow constraints that connect model outputs
WIRED
ICE’s New Detention Center Contracts Declare State Laws ‘Shall Not Apply’
・One day after a federal judge ordered an ICE detention center opened to state health inspectors, the agency posted new contract terms that would void state oversight at four facilities.
cs.LG updates on arXiv.org
IKS-Instruct: A 24,000-Example Multilingual Dataset for Teaching Language Models Indian Knowledge Systems
・arXiv:2607.23322v1 Announce Type: cross Abstract: Instruction tuning has become the standard method for adapting large language models to follow human intent, yet existing instruction datasets are dominated by English-language general-knowledge tasks and lack coverage of specialized pedagogical domains.
・This paper presents IKS-Instruct, a dataset of 24,795 instruction-response pairs for teaching language models to de
cs.LG updates on arXiv.org
Infinite-Precision Autoregressive Modeling for Vector Graphics and Layouts
・arXiv:2601.05680v2 Announce Type: replace Abstract: While Transformer-based autoregressive models excel in data generation, their token discretization strategy inherently limits their precision in continuous domains.
・We analyze the scalability limitations of existing discretization-based approaches for generating hybrid discrete-continuous sequences, particularly in high-precision domains such as logos, layouts, and
cs.LG updates on arXiv.org
INSIGHT: Spatially resolved survival modelling from routine histology crosslinked with molecular profiling reveals prognostic epithelial-immune axes in stage II/III colorectal cancer
・arXiv:2512.22262v2 Announce Type: replace-cross Abstract: Routine histology contains rich prognostic information in stage II/III colorectal cancer, much of which is embedded in complex spatial tissue organisation.
・We present INSIGHT, a graph neural network that predicts survival directly from routine histology images.
・Trained and cross-validated on TCGA (n=342) and SURGEN (n=336), INSIGHT produces patient-level spati
cs.LG updates on arXiv.org
Investigating the Visual Cues of CNNs for Vascular Segmentation: A Case Study in Microscopy and Fundus Imaging
・arXiv:2607.23371v1 Announce Type: cross Abstract: Vascular segmentation is a standard procedure for clinical diagnosis, yet the specific visual features determining model decisions remain poorly understood.
・This paper investigates the visual cues Convolutional Neural Networks (CNNs) use to segment blood vessels across two distinct imaging domains: fluorescence microscopy and retinal fundus photography.
・We employ a se
cs.LG updates on arXiv.org
Kan Extension Transformers: A Categorical Unification of Attention, Diffusion, and Predict-Detach Self-Conditioning
・arXiv:2605.27259v2 Announce Type: replace Abstract: We propose Kan Extension Transformers (KETs) as a categorical design language for a diverse group of Transformer implementations.
・A layer can be viewed generally as a weighted structured extension operator: attention uses token neighborhoods, geometric mixing uses sparse incidences, and KET uses simplicial sources.
・This operator is an actual enriched left Kan extens
Hugging Face Papers
Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory
Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory
cs.LG updates on arXiv.org
Key-Value Means: Transformers with Expandable Block-Recurrent Compressed Memory
・arXiv:2605.09877v4 Announce Type: replace Abstract: Recall presents a difficult choice: transformers have a linearly growing memory that slows each successive token, while linear RNNs typically have fixed costs but limited recall.
・We present Key-Value Means ("KVM"), a novel block-recurrence for attention that can accommodate either fixed-size or growing state.
・Equipping a strong transformer baseline with fixed-size K
cs.LG updates on arXiv.org
Kimi K3: Open Frontier Intelligence
・arXiv:2607.24653v1 Announce Type: cross Abstract: We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window.
・Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth.
・Together with Stable LatentMoE, which effectively activates 16 o
Zennの「大規模言語モデル」のフィード
Kimi K3は何が公開されたのか：2.8兆パラメータと約1.56TBの意味
・Kimi K3は、Moonshot AIが公開した2.8兆パラメータの大規模言語モデルです。
・数字だけでもかなり目を引きますが、個人的に面白いと感じたのは、推論に使えるフルウェイトまでHugging Faceで公開されたことでした。公式リポジトリ全体は、2026年7月29日時点で約1.56TBあります。
・ただし、ここで混同したくないのが次の2点です。
cs.LG updates on arXiv.org
LanteRn: Latent Visual Structured Reasoning
・arXiv:2603.25629v2 Announce Type: replace-cross Abstract: While language reasoning models excel in many tasks, visual reasoning remains challenging for current large multimodal models (LMMs).
・As a result, most LMMs default to verbalizing perceptual content into text, a strong limitation for tasks requiring fine-grained spatial and visual understanding.
・While recent approaches take steps toward thinking with images by
cs.LG updates on arXiv.org
Latent Confounded Causal Discovery via Lie Bracket Geometry
・arXiv:2606.19610v2 Announce Type: replace Abstract: We study causal discovery from observational and interventional regimes when latent variables may affect the measured system.
・Our first algorithm, BRIDGE (Bracket Residuals for Interventional Discovery and Geometric Estimation), combines a density-ratio or transport engine with a high-recall geometric screen and passes the retained arrows to a score-based or differe
cs.LG updates on arXiv.org
Learning Asymptotics with Convergence-Rate Guarantees using Linear Least Squares
・arXiv:2607.23287v1 Announce Type: cross Abstract: We introduce a new research area that is called Asymptotics Learning Theory (ALT) and combines optimization with asymptotic analysis.
・In particular, ALT provides a unified approach for computing unknown constants/parameters in proven asymptotic expansions using optimization theory.
・In this paper, we focus on a general asymptotic form which includes a broad class of as
cs.LG updates on arXiv.org
Learning Distributions from Multiple Data Providers
・arXiv:2607.24732v1 Announce Type: cross Abstract: Motivated by learning from heterogeneous and overlapping data providers, we study a stylized model of distribution learning from restricted conditional samples.
・The goal is to learn an unknown distribution $p$ on a finite domain $[n]$.
・The learner is given a fixed family of queryable sets $\mathscr{S} \subseteq 2^{[n]}$, and each query to $S \in \mathscr{S}$ returns a
stat.ML updates on arXiv.org
Learning from the Unseen: Offline Reinforcement Learning with Hidden Actions
・arXiv:2607.25241v1 Announce Type: new Abstract: Standard offline reinforcement learning (RL) algorithms typically assume that the actions in the dataset are observed without error.
・However, in many real-world applications, the true actions are unobserved and only noisy proxies are available, causing existing RL methods to yield biased and potentially misleading conclusions.
・We study off-policy evaluation in infinite-
cs.LG updates on arXiv.org
Learning switched non-linear dynamical systems from a single trajectory
・arXiv:2607.23502v1 Announce Type: cross Abstract: We study empirical risk minimization for learning non-linear dynamical systems whose transition dynamics may switch over time.
・Under stability assumptions, and i.i.d switching over a set of $K$ modes, we derive non-asymptotic bounds on the prediction risk expressed in terms of the metric entropy of the underlying function class.
・We instantiate our general result for H
cs.LG updates on arXiv.org
LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models
・arXiv:2411.00918v5 Announce Type: replace-cross Abstract: Mixture of experts (MoE) architectures have become a cornerstone for scaling up and are a key component in most large language models such as GPT-OSS, DeepSeek-V3, Llama-4, and Gemini-2.5.
・However, systematic research on MoE remains severely constrained by the prohibitive computational costs of training and evaluation, restricting large-scale studies accessibl
MarkTechPost
Liquid AI Releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: Bidirectional Encoders That Stay Fast at 8K Context on CPU
・Liquid AI released two open-weight bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M.
・Both carry an 8,192-token context and are built on the LFM2 hybrid backbone.
・The 350M ranks fourth of 14 models on a 17-task GLUE, SuperGLUE, and multilingual suite, behind only larger models.
Zennの「大規模言語モデル」のフィード
LLM モデルの差し替えを自動化する llm-replacer の紹介
・はじめに こんにちは、PKSHA Technology の深堀です。
・LLM を組み込んだサービスでは、モデルが EOL（End of Life）を迎えた際にモデル更新が必要です。
・ただ、実際に運用中サービスのモデルを差し替えようとすると、単に model_id を 1 行変えれば終わり、とはなかなかなりません。
cs.LG updates on arXiv.org
LLM-Based Scientific Equation Discovery via Physics-Informed Token-Regularized Policy Optimization
・arXiv:2602.10576v2 Announce Type: replace Abstract: Symbolic regression aims to distill mathematical equations from observational data.
・Recent approaches have successfully leveraged Large Language Models (LLMs) to generate equation hypotheses, capitalizing on their vast pre-trained scientific priors.
・However, existing frameworks predominantly treat the LLM as a static generator, relying on prompt-level guidance to st
cs.LG updates on arXiv.org
LLM-based Source Code Compression via Thresholded Symbol Ranking
・arXiv:2607.24192v1 Announce Type: cross Abstract: We study the problem of lossless compression of source code, motivated by the storage demands of large-scale software archives, such as Software Heritage (https://www.softwareheritage.org/).
・General-purpose compressors (e.g., zstd, bzip2) offer a good trade-off between compression ratio and speed, but fail to exploit all special regularities inherent in source code.
#LLMタグ
LLM【大規模言語モデル】９社比較してみた。
・【2026年最新版】主要AI 9社を徹底比較 ChatGPT・Claude・Gemini・Grok・Kimi・Qwen・Copilot・DeepSeek・Meta・Sakana Fuguの違いとは？ 「結局、どのAIを使えばいいの？」 続きをみる
stat.ML updates on arXiv.org
Lloyd's $K$-Means Clustering Algorithm Is Frank-Wolfe in Disguise
・arXiv:2607.25190v1 Announce Type: new Abstract: Lloyd's $K$-means algorithm, also known as na\"{i}ve $K$-means, is a widely used ad hoc optimization heuristic, designed to minimize the sum of squared errors (SSE) across all $K$-partitions of a dataset via iterative cluster refinement.
・In this work, we establish a novel connection between Lloyd's algorithm and the Frank-Wolfe (FW) algorithm, a prominent first-order me
Zennの「大規模言語モデル」のフィード
Local LLM で画像の PII マスキングを試してみた
・TL;DR ローカルLLMを使ったPIIマスキングの検証を行いました LLMの適用箇所をテキストに限定することで、実用的な速度で処理できることがわかりました 1.
・はじめに Finatextでは5月にAIコンテストが行われ、私の参加したチームでは動画からマニュアルつくるくんを作りました。
・詳しくは以下の記事をご覧ください。
cs.LG updates on arXiv.org
Logit-Coordinate Generative Models for Mixed Continuous-Categorical Tabular Data
・arXiv:2607.23348v1 Announce Type: cross Abstract: Mixed continuous--categorical data pose a representation problem for continuous generative models.
・Flow Matching and Gaussian diffusion operate in Euclidean spaces, whereas categorical laws lie on probability simplices and may be highly imbalanced.
・We study a logit-coordinate framework that encodes categorical variables as smoothed natural parameters and combines them
cs.LG updates on arXiv.org
Long-Tailed Medical Image Classification
・arXiv:2607.23883v1 Announce Type: cross Abstract: In this paper, we examine the difficulties of using standard techniques for medical image classification due to long-tailed distributions (wherein rarer conditions have very few samples) resulting in bias towards diagnosing common diseases and away from rarer diseases.
・We then discuss and implement deep learning models with techniques such as augmentation to minimize
cs.LG updates on arXiv.org
Loong: Synthesize Long Chain-of-Thoughts at Scale through Verifiers
・arXiv:2509.03059v2 Announce Type: replace Abstract: Recent advances in Large Language Models (LLMs) have shown that their reasoning capabilities can be significantly improved through Reinforcement Learning with Verifiable Reward (RLVR), particularly in domains like mathematics and programming, where ground-truth correctness can be automatically evaluated.
・However, extending this success to other reasoning-intensive d
WIRED
Love It or Hate It, the Ferrari Luce Is a Thrilling Drive
・WIRED test-drove a preproduction version of the Luce, the storied Italian automaker’s controversial electric car, and it was remarkably surprising.
・Will it convert Ferrari fanatics?
WIRED
Mac Mini Availability: Long Waits and Higher Prices
・Thanks to the surge in local AI processing and the ongoing memory shortage, it’s incredibly difficult to buy a Mac Mini.
cs.LG updates on arXiv.org
Machine Learning for Cloud Detection in IASI Measurements: A Data-Driven SVM Approach with Physical Constraints
・arXiv:2508.10120v2 Announce Type: replace-cross Abstract: Cloud detection is fundamental for the interpretation and operational exploitation of hyperspectral infrared sounders, yet the capability of infrared radiances alone to provide reliable cloud information remains insufficiently assessed.
・We introduce the Cloud Identification Support Vector Machine (CISVM), a supervised framework for global clear and cloudy clas
Hugging Face Papers
Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model
Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model
Hugging Face Papers
Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion
Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion
cs.LG updates on arXiv.org
Market Design for AI: Beyond the Copyright Binary
・arXiv:2606.12260v2 Announce Type: replace-cross Abstract: How can we design a market of human-generated content for use in training AI models that both enables technological progress and preserves individual incentives for high-quality content creation?
・Existing approaches take polar positions: a "free-for-all" model based on fair use and a "strong intellectual property rights" model.
・We show that both fail: Free-for
cs.LG updates on arXiv.org
MaskAttn-SDXL: Controllable Region-Level Text-To-Image Generation
・arXiv:2509.15357v3 Announce Type: replace-cross Abstract: Diffusion models have achieved strong results in text-to-image generation, but important limitations remain as prompts become more structured and multi-object.
・On the architecture side, U-Net backbones are efficient and stable, yet their locality makes global coordination harder, while Transformer-based diffusion models improve global interactions but at subst
ITmedia NEWS 最新記事一覧
Meta、エルパソのデータセンターをBlackRockと共同開発 総額140億ドル、Metaの出資比率は20％
・Metaは、BlackRockとテキサス州エルパソのAIデータセンターキャンパスを開発・保有する合弁事業を設立すると発表した。総開発費約140億ドルのうちBlackRock側が80％を保有し、Metaは完成後にキャンパスを賃借する。外部資本を活用してバランスシートの負担を抑えつつ計算基盤を拡大する狙いだ。
cs.LG updates on arXiv.org
Minimax Lower Bounds of Kernel Discrepancy Estimation: MMD, HSIC, KSD
・arXiv:2607.24235v1 Announce Type: cross Abstract: Over the past 20 years, kernel discrepancies have been leveraged as a highly powerful tool for quantifying the disagreement of distributions, with numerous successful applications in two-sample, goodness-of-fit, and independence testing, among others.
・Their fastest estimators are known to converge at a parametric rate---$n^{-1/2}$---under mild conditions.
・While this r
cs.LG updates on arXiv.org
MMOE: Modernizing Diffusion Transformers with Efficient Expert Design
・arXiv:2607.24665v1 Announce Type: cross Abstract: Modern large language models scale successfully by pairing capacity growth with efficiency, keeping per-token and deployment costs under control as capacity grows.
・AIGC Foundation Models (AFMs), especially diffusion-transformer backbones, have begun to adopt sparse experts, but recent efforts mostly enlarge total parameter counts and sparsity ratios without importing
cs.LG updates on arXiv.org
MOCA: A Transformer-based Modular Causal Inference Framework with One-way Cross-attention and Cutting Feedback
・arXiv:2604.23107v2 Announce Type: replace-cross Abstract: Causal effect estimation from observational data requires careful adjustment for confounding.
・Classical estimators such as inverse probability weighting and augmented inverse probability weighting can perform well under favorable model specification but may become unstable in complex settings.
・Machine-learning and representation-learning methods provide greate
Hugging Face Papers
MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities
MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities
cs.LG updates on arXiv.org
MolCryst-MLIPs: A Machine-Learned Interatomic Potentials Database for Molecular Crystals
・arXiv:2604.13897v2 Announce Type: replace Abstract: We present an open Molecular Crystal (MC) database of Machine-Learned Interatomic Potentials (MLIP) called MolCryst-MLIPs.
・The first release comprises fine-tuned MACE models for nine molecular crystal systems---Benzamide, Benzoic acid, Coumarin, Durene, Isonicotinamide, Nicotinic acid , Niacinamide, Pyrazinamide, and Resorcinol---developed using the Automated Machin
WIRED
More Typos, Fewer Em Dashes: Writers Are Creating an Anti-AI ‘Literary Counterculture’
・Novelists, journalists, and power LinkedIn posters are embracing first-person narratives and idiosyncrasies to avoid being mistaken for chat bots.
cs.LG updates on arXiv.org
Multi-Task GRPO: Reliable LLM Reasoning Across Tasks
・arXiv:2602.05547v2 Announce Type: replace-cross Abstract: RL-based post-training with GRPO is widely used to improve large language models on individual reasoning tasks.
・However, real-world deployment requires reliable performance across diverse tasks.
・A straightforward multi-task adaptation of GRPO often leads to imbalanced outcomes, with some tasks dominating optimization while others stagnate.
stat.ML updates on arXiv.org
Multiclass Classification without Labels via Posterior Simplex Geometry
・arXiv:2607.24943v1 Announce Type: cross Abstract: In many classification problems, reliable instance-level labels are unavailable.
・However, it is often possible to construct weakly enriched unlabeled samples: datasets selected by different cuts, sources, populations, or experimental conditions that change latent class proportions without revealing them.
・Classification without Labels (CWoLa) shows that, in the binary
cs.LG updates on arXiv.org
Multivariate Time Series Forecasting with Adaptive Non-Local Observables
・arXiv:2607.24399v1 Announce Type: cross Abstract: Multivariate time series forecasting (MTSF) predicts future values of multiple variables from historical data.
・While quantum neural networks have been increasingly applied to this task, they typically rely on fixed local measurements, which restrict their expressivity.
・We propose MTSF-ANO, a simple hybrid model for MTSF that integrates variational quantum circuits wit
cs.LG updates on arXiv.org
Music-Source-Separation-Training (MSST): A Unified Framework for Training and Evaluating Music Demixing Models
・arXiv:2607.23395v1 Announce Type: cross Abstract: Music Source Separation (MSS), the task of recovering individual sound components (stems) from a polyphonic mixture, is central to applications ranging from karaoke and remixing to audio restoration and content production.
・The separation quality depends on engineering decisions across the entire pipeline: model choice, training data preparation and augmentation, loss
WIRED
NASA’s New 3D Model Shows the Earth Is a Lumpy Mess
・We like to think of our home as a nice, smooth sphere.
・But mapping the Earth’s gravitational field provides a different view of the planet.
cs.LG updates on arXiv.org
Nearly Tight Bounds for Cross-Learning Contextual Bandits with Graphical Feedback
・arXiv:2502.04678v2 Announce Type: replace Abstract: Repeated first-price auctions are contextual decision problems with censored but reusable feedback: after submitting a bid, a learner can infer the outcomes of related bids and evaluate them under different private values.
・This structure motivates cross-learning contextual bandits with graphical feedback, where playing an arm reveals the losses of its out-neighbors
cs.LG updates on arXiv.org
Neural Message-Passing on Attention Graphs for Hallucination Detection
・arXiv:2509.24770v2 Announce Type: replace Abstract: Large Language Models (LLMs) often generate incorrect or unsupported content, known as hallucinations.
・Existing detection methods rely on heuristics or simple models over isolated computational traces such as activations, or attention maps.
・We unify these signals by representing them as attributed graphs, where tokens are nodes, edges follow attentional flows, and b
stat.ML updates on arXiv.org
Neural Networks Provably Learn Spectral Representations for Group Composition
・arXiv:2606.02993v2 Announce Type: replace-cross Abstract: Understanding how structured internal structure emerges during neural network training is central to the study of deep learning.
・We investigate this phenomenon through the group composition task, where a two-layer neural network is trained to predict $g_1 \star g_2$ for elements of a finite group $G$.
・By lifting the projected gradient flow to the Fourier domai
cs.LG updates on arXiv.org
Neural Representation of Minimal Surfaces
・arXiv:2607.23437v1 Announce Type: cross Abstract: We propose a neural representation for minimal surfaces.
・Unlike prior approaches based on discretization or Physics-Informed Neural Networks (PINNs), where meshes or neural fields are optimized to approximate the governing equations, our method builds on an exact representation, similar to the classical Weierstrass--Enneper parameterization, yielding minimal surfaces
cs.LG updates on arXiv.org
No Free Lunch in Flow Surrogates under Time-Varying Boundary Conditions: A Two-Regime Study
・arXiv:2607.23667v1 Announce Type: cross Abstract: A flow surrogate validated on a simple regime is often taken as evidence that the approach will carry to a richer one.
・We test this assumption on two transient flows under time-varying boundary conditions emulating the process startup: the three-dimensional slurry film in chemical-mechanical planarisation (CMP), a core semiconductor-manufacturing process, and the two-
cs.LG updates on arXiv.org
No Optimal Language Set Exists for Multilingual Instruction Tuning: Insights from a Linguistically-Informed Study
・arXiv:2410.07809v2 Announce Type: replace-cross Abstract: Multilingual instruction tuning (MIT) is challenged by the curse of multilinguality, data scarcity, and high computational cost.
・A natural hypothesis is that carefully selecting a linguistically diverse set of languages yields universally better models.
・We test this systematically by evaluating linguistically-informed selection strategies---based on typologica
cs.LG updates on arXiv.org
NormGuard: Reward-Preserving Norm Constraints in Flow-Matching Reinforcement Learning
・arXiv:2606.27771v3 Announce Type: replace Abstract: Reinforcement learning (RL) post-training improves the reward alignment of flow-based generators, but often degrades perceptual quality in ways that are not captured by the reward proxy.
・We identify a simple structural signature of this drift: across three post-training methods (NFT, AWM, DPO), RL fine-tuning inflates the per-step velocity norm $\|v_\theta\|$ by $5\
Hugging Face Papers
Novel Claim or Déjà Vu? Rethinking "Contamination-Free'' Dynamic Evaluation for Multimodal Automated Fact-Checking
Novel Claim or Déjà Vu? Rethinking "Contamination-Free'' Dynamic Evaluation for Multimodal Automated Fact-Checking
cs.LG updates on arXiv.org
Numerical Investigation of Sequence Modeling Theory using Controllable Memory Functions
・arXiv:2506.05678v3 Announce Type: replace Abstract: The evolution of sequence modeling architectures, from recurrent neural networks and convolutional models to Transformers and structured state-space models, reflects ongoing efforts to address the diverse temporal dependencies inherent in sequential data.
・Despite this progress, systematically characterizing the strengths and limitations of these architectures remain
cs.LG updates on arXiv.org
Offline-to-Online Creative Optimization with Generative Models and Adaptive Testing
・arXiv:2607.23696v1 Announce Type: cross Abstract: Ad creative optimization is increasingly constrained by evaluation rather than generation.
・Generative models can produce many plausible creatives, but reliable evaluation requires online experiments, in which only a limited slate can be tested.
・We study how to use data from historical A/B tests to generate and select the candidates in that slate.
Hugging Face Papers
OmniDelta: Skill-Driven Budget Allocation for Token Compression in OmniLLMs
OmniDelta: Skill-Driven Budget Allocation for Token Compression in OmniLLMs
cs.LG updates on arXiv.org
On a linear fused Gromov-Wasserstein distance for graph structured data
・arXiv:2203.04711v2 Announce Type: replace Abstract: We present a framework for embedding graph structured data into a vector space, taking into account node features and topology of a graph into the optimal transport (OT) problem.
・Then we propose a novel distance between two graphs, named linearFGW, defined as the Euclidean distance between their embeddings.
・The advantages of the proposed distance are twofold: 1) it
cs.LG updates on arXiv.org
On Non-Stationary Dynamic Pricing: Adaptivity and Optimality
・arXiv:2607.24115v1 Announce Type: cross Abstract: We study the contextual dynamic pricing problem under non-stationarity, where a firm sells products to $T$ sequentially arriving consumers that behave according to an unknown demand model that can change over time.
・The demand model is assumed to be a generalized linear model (GLM), allowing for a feature vector in $\mathbb{R}^d$ that encodes products and consumer info
stat.ML updates on arXiv.org
On the Convergence Analysis of Muon
・arXiv:2505.23737v3 Announce Type: replace Abstract: The majority of parameters in neural networks are naturally represented as matrices.
・However, most commonly used optimizers treat these matrix parameters as flattened vectors during optimization, potentially overlooking their inherent structural properties.
・Recently, an optimizer called Muon has been proposed, specifically designed to optimize matrix-structured para
cs.LG updates on arXiv.org
On the Detection of Commutative Factors in Factor Graphs: Necessary and Sufficient Conditions
・arXiv:2605.26908v2 Announce Type: replace-cross Abstract: Exploiting the indistinguishability of objects in a probabilistic graphical model such as a factor graph is key to lifted probabilistic inference algorithms and allows for tractable probabilistic inference problems with respect to domain sizes.
・A central building block for the exploitation of indistinguishable objects in factor graphs is the identification of
cs.LG updates on arXiv.org
On-Device Inference versus Wireless Streaming: Energy-Efficient Multi-Modal Deep Learning for Wearable Cardiovascular Patches
・arXiv:2510.18668v4 Announce Type: replace Abstract: Wearable cardiovascular sensor patches promise continuous, unobtrusive monitoring, but their tight energy, memory, and compute budgets make it unclear whether physiological signals should be analyzed on the device or streamed to the cloud for processing.
・We study this inference-versus-transmission trade-off for a resource-constrained patch that records synchronized
cs.LG updates on arXiv.org
Online Fair Division with Budget Constraints
・arXiv:2607.23310v1 Announce Type: cross Abstract: We study an online variant of discrete fair division under generalized assignment budget constraints.
・Goods arrive one at a time and must be assigned irrevocably to a feasible agent or to charity, which holds all unallocated goods, while fairness is evaluated only against budget-feasible subsets of every recipient's bundle.
・We first show that, without additional struc
The Verge
OpenAI’s rogue AI agent didn’t stop at hacking Hugging Face
・The AI agent that escaped from OpenAI and hacked developer platform Hugging Face attacked other companies as well, OpenAI revealed on Tuesday.
・The update substantially widens the scope of an already concerning incident, which has alarmed industry insiders and fueled growing calls for stronger oversight on frontier AI systems.
・In an update to a blog post detailing its ongoing investigation into the incident, OpenAI sa
#LLMタグ
OpenAIがひっそりOSS化した「Codex Security」が、セキュリティスキャンの常識を変えようとしている話
・今回は、OpenAIが表立った発表もなくオープンソース公開したセキュリティスキャナ「Codex Security」について、調べたことを整理していきます。
ITmedia NEWS 最新記事一覧
OpenAIの暴走AIエージェント、別企業のシステムにも不正侵入
・OpenAIの隔離環境を脱出したAIエージェントが、Hugging Faceへのサイバー攻撃に加え、Modal Labsの顧客システムにも侵入していたことが分かった。
Zennのトレンド
OpenTelemetry初心者がCollectorの中身を全部分解して理解してみた
・OpenTelemetry Collectorは何をしているのか？ Kubernetes上でReceiver・Processor・Exporterを徹底的に追いかけてみた この記事のポイント OpenTelemetry Collectorを導入すると、 Receiver Processor Exporter Pipeline といった概念が登場します。
・私は当初、Collectorを「データを転送するためのコンポーネント」程度に捉えていました。
・しかし実際にKubernetes環境で、 メトリクス トレース ログ を収集しながら内部構造を追っていくと、Collecto...
Hugging Face Papers
OPERA: Offline Policy-guided Expert Routing and Adaptation for Universal Biomedical Image Analysis
OPERA: Offline Policy-guided Expert Routing and Adaptation for Universal Biomedical Image Analysis
cs.LG updates on arXiv.org
Operator learning for models of tear film breakup
・arXiv:2601.08001v2 Announce Type: replace-cross Abstract: Tear film (TF) breakup is a key driver of understanding dry eye disease, yet estimating TF thickness and osmolarity from fluorescence (FL) imaging typically requires solving computationally expensive inverse problems.
・We propose an operator learning framework that replaces traditional inverse solvers with neural operators trained on simulated TF dynamics.
cs.LG updates on arXiv.org
Order in Desbordante: Techniques for Efficient Implementation of Order Dependency Discovery Algorithms
・arXiv:2607.23632v1 Announce Type: cross Abstract: Science-intensive data profiling focuses on discovery and validation of various patterns in datasets.
・This study considers discovery of one such pattern - order dependency (OD).
・Simply put, OD states that some list of columns is ordered according to another one.
cs.LG updates on arXiv.org
Ordering-based Causal Discovery via Generalized Score Matching
・arXiv:2601.16249v3 Announce Type: replace Abstract: Learning DAG structures from purely observational data remains a long-standing challenge across scientific domains.
・An emerging line of research leverages the score of the data distribution to initially identify a topological order of the underlying DAG via leaf node detection and subsequently performs edge pruning for graph recovery.
・This paper extends the score ma
Hugging Face Papers
Parallel Decoding Distillation for Fast Image and Video Generation
Parallel Decoding Distillation for Fast Image and Video Generation
cs.LG updates on arXiv.org
Parallel Swin Transformer-Enhanced 3D MRI-to-CT Synthesis for MRI-Only Radiotherapy Planning
・arXiv:2602.05387v2 Announce Type: replace-cross Abstract: MRI provides superior soft tissue contrast without ionizing radiation; however, the absence of electron density information limits its direct use for dose calculation.
・As a result, current radiotherapy workflows rely on combined MRI and CT acquisitions, increasing registration uncertainty and procedural complexity.
・Synthetic CT generation enables MRI only plan
Hugging Face Papers
Pass the Baton: Trajectory-Relayed On-Policy Distillation
Pass the Baton: Trajectory-Relayed On-Policy Distillation
cs.LG updates on arXiv.org
PathRIR: Physics-Guided Acoustic Path Selection and Late-Tail Compensation for Fast Room Impulse Response Simulation
・arXiv:2607.23293v1 Announce Type: cross Abstract: Image-source-method (ISM)-based room impulse response (RIR) simulation is a useful and physically interpretable tool for acoustic scene modeling, but full-order ISM becomes computationally expensive as the reflection order and room complexity increase.
・We propose a physics-guided framework for fast RIR simulation that preserves the geometric structure of ISM while lea
ITmedia NEWS 最新記事一覧
PayPayアプリで熊本地震への寄付が可能に 最短3ステップ・1円から
・PayPay社は7月29日、決済アプリ「PayPay」に寄付機能を追加した。28日に発生した令和8年熊本地震の緊急支援募金にも寄付できる。
cs.LG updates on arXiv.org
PD$^3$: A Project Duplication Detection Framework via Adapted Multi-Agent Debate
・arXiv:2505.17492v2 Announce Type: replace-cross Abstract: Project duplication detection is critical for project quality assessment because it helps avoid investment in repeated proposals.
・Existing methods usually cast it as ranking and rely on surface matching or direct large language models judging, often missing practical needs in set-level reference selection.
・We recast the task as many-to-many reference set selec
cs.LG updates on arXiv.org
PeopleSearchBench: A Multi-Dimensional Benchmark for Evaluating AI-Powered People Search Platforms
・arXiv:2603.27476v2 Announce Type: replace-cross Abstract: AI-powered people search platforms are increasingly used in recruiting, sales prospecting, and professional networking, yet no widely accepted benchmark exists for evaluating their performance.
・We introduce PeopleSearchBench, an open-source benchmark that compares four people search platforms on 119 real-world queries across four use cases: corporate recruitin
Hugging Face Papers
PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models
PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models
cs.LG updates on arXiv.org
Photonic reservoir computing with complex networks
・arXiv:2607.23285v1 Announce Type: cross Abstract: Photonic reservoir computing has attracted increasing attention as a fast and low-cost approach for time-series prediction.
・Photonic reservoir computing utilizes the high speed, broad bandwidth, and spatial parallelism of light.
・However, the effect of the internal connection structure (network topology) on the computing performance has not been investigated for large-
cs.LG updates on arXiv.org
Physics-Encoded Inverse Modeling for Arctic Snow Depth Estimation
・arXiv:2601.17074v5 Announce Type: replace Abstract: Accurate estimation of unobserved quantities in time-varying inverse problems remains challenging when observations are sparse and only indirectly related to the target variable.
・In Arctic climate applications, snow depth over sea ice is not directly available in commonly used reanalysis products and must instead be inferred from related physical and environmental v
cs.LG updates on arXiv.org
Plain Transformers are Surprisingly Powerful Link Predictors
・arXiv:2602.01553v3 Announce Type: replace Abstract: Link prediction is a core challenge in graph machine learning, demanding models that capture rich and complex topological dependencies.
・While Graph Neural Networks (GNNs) are the standard solution, state-of-the-art pipelines often rely on explicit structural heuristics or memory-intensive node embeddings -- approaches that struggle to generalize or scale to massive
#LLMタグ
Preventing Misinterpretation of NRA-IDE
・Canonical English Comments, Reference Code, and Deterministic Tests 続きをみる
cs.LG updates on arXiv.org
Principles and Guidelines for Randomized Controlled Trials in AI Evaluation
・arXiv:2605.02050v2 Announce Type: replace-cross Abstract: This work establishes a framework for standardizing AI evaluation RCTs (sometimes called human uplift studies).
・Drawing on established practices from disciplines with established RCT traditions, including software engineering, economics, clinical and health sciences, and psychology, we synthesize five principles drawn from established validity frameworks and o
cs.LG updates on arXiv.org
Procedural Content Generation via Generative Artificial Intelligence
・arXiv:2407.09013v2 Announce Type: replace-cross Abstract: The attempt to utilize machine learning in PCG has been made in the past.
・In this survey paper, we investigate how generative artificial intelligence (AI), which saw a significant increase in interest in the mid-2010s, is being used for PCG.
・We review applications of generative AI for the creation of various types of content, including terrains, items, and eve
Hugging Face Papers
Projection Pursuit CPCANet for Domain Generalization
Projection Pursuit CPCANet for Domain Generalization
cs.LG updates on arXiv.org
proxymate: Diagnosis and Adjustment of Proxy Estimates for Reliable Inference
・arXiv:2607.24401v1 Announce Type: cross Abstract: Proxy outcomes (such as short-term behavioral signals, model predictions, or surrogate endpoints) are frequently used in place of primary outcomes that are too slow to mature, rare, or challenging to measure directly.
・But valid inference on a proxy does not guarantee valid inference on the primary estimate as proxy-based estimates can be systematically biased in ways
cs.LG updates on arXiv.org
PS-PPO: Prefix-Sampling PPO for Critic-Free RLHF
・arXiv:2606.29758v2 Announce Type: replace Abstract: Reinforcement Learning from Human Feedback (RLHF) for Large Language Models increasingly relies on critic-free methods as a practical alternative to actor--critic training.
・Despite their simplicity, existing critic-free approaches propagate a trajectory-level learning signal uniformly across all tokens in a trajectory.
・This requires full-trajectory policy updates fo
機械学習タグが付けられた新着記事 - Qiita
Python・NumPy・PyTorchの乱数seedをまとめて固定する
・機械学習やシミュレーションでは、同じコードを実行しているのに結果が毎回変わることがあります。
・原因の一つが、Python・NumPy・PyTorchで別々の乱数生成器が使われていることです。
・そこで、各ライブラリのseedをまとめて固定する関数を作ります。
cs.LG updates on arXiv.org
Quantum Safe-Set Bayesian Optimization for Quality Improvement in Fuselage Assembly
・arXiv:2511.22090v2 Announce Type: replace Abstract: Recent efforts in smart manufacturing have enhanced aerospace fuselage assembly processes, particularly by innovating shape adjustment techniques to minimize dimensional gaps between assembled sections.
・Existing approaches have shown promising results but face the issue of low sample efficiency from the manufacturing systems.
・It arises from the limitation of the cla
cs.LG updates on arXiv.org
Realizing Scaling Laws in Recommender Systems: A Foundation-Expert Paradigm for Hyperscale Model Deployment
・arXiv:2508.02929v3 Announce Type: replace-cross Abstract: Scaling laws have been established for recommender systems, yet efficiently deploying foundation model (FM) across multiple recommendation surfaces remains a major unsolved challenge.
・Existing methods for transfer learning face fundamental limitations in this setting: knowledge distillation suffers from transfer fidelity degradation in the large-data regime, a
cs.LG updates on arXiv.org
Reconstruction of SINR Maps from Sparse Measurements using Group Equivariant Non-Expansive Operators
・arXiv:2507.19349v3 Announce Type: replace Abstract: As sixth generation (6G) wireless networks evolve, accurate signal-to-interference-noise ratio (SINR) maps are becoming increasingly critical for effective resource management and optimization.
・However, acquiring such maps at high resolution is often cost-prohibitive, creating a severe data scarcity challenge.
・This necessitates machine learning (ML) approaches capab
Hugging Face Papers
ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition
ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition
stat.ML updates on arXiv.org
Regime-Adaptive Bayesian Optimization via Dirichlet Process Mixtures of Gaussian Processes
・arXiv:2601.20043v2 Announce Type: replace-cross Abstract: Standard Bayesian Optimization (BO) assumes uniform smoothness across the search space an assumption violated in multi-regime problems such as molecular conformation search through distinct energy basins or drug discovery across heterogeneous molecular scaffolds.
・A single GP either oversmooths sharp transitions or hallucinates noise in smooth regions, yielding
Hugging Face Papers
Reinforcement Learning for Code Optimization
Reinforcement Learning for Code Optimization
stat.ML updates on arXiv.org
Reinformed Dreamer: An Asymmetric World Model Efficiently Trained through Latent Guidance
・arXiv:2607.26040v1 Announce Type: cross Abstract: Much like humans benefit from guidance while learning, reinforcement learning algorithms may benefit from additional supervision beyond rewards.
・Leveraging additional information during training to learn better representations and behaviors has been the focus of asymmetric reinforcement learning.
・This learning paradigm has proven effective under partial observability
The Verge
Remarkable’s refurbished bundle is an awesome deal that’s over $350 off
・The Remarkable Paper Pro bundle comes with a stylus and a folio cover.
・A lot of us here at The Verge are fans of Remarkable’s digital note-taking tablets, and right now one of its higher-end options is much cheaper than usual.
・The Remarkable Paper Pro normally sells for $629 on its own, but Woot is currently offering a refurbished bundle for $449.99 through July 31st.
cs.LG updates on arXiv.org
Rendering on Real Silicon: GPU Render-Timing as a Passive, AI-Resistant CAPTCHA Signal
・arXiv:2607.23389v1 Announce Type: cross Abstract: Conventional CAPTCHAs pose puzzles that modern AI systems increasingly solve, while behavioral and cryptographic-attestation defenses carry privacy or enrollment costs.
・We investigate an orthogonal signal: the physical timing behavior of a client's GPU under a controlled WebGL rendering workload.
・Unlike WebGL fingerprinting, which hashes pixel output into a static dev
cs.LG updates on arXiv.org
Repeated-Token Counting Reveals a Dissociation Between Representations and Outputs
・arXiv:2605.09239v2 Announce Type: replace-cross Abstract: Large language models fail at counting how many times a word repeats in a list, even though they perform well on far harder reasoning tasks.
・These failures are commonly attributed to limitations in internal count tracking.
・We show this attribution is wrong.
cs.LG updates on arXiv.org
Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
・arXiv:2607.24731v1 Announce Type: cross Abstract: On-policy distillation (OPD) adapts diffusion models by querying a teacher along trajectories generated by the current student, but how it should behave under classifier-free guidance (CFG), a default component of modern diffusion systems, remains poorly understood.
・Existing OPD methods naturally extend velocity matching to the CFG-composed prediction, directly matchi
cs.LG updates on arXiv.org
Reverso: Efficient Time Series Foundation Models for Zero-shot Forecasting
・arXiv:2602.17634v2 Announce Type: replace Abstract: Learning time series foundation models has been shown to be a promising approach for zero-shot time series forecasting across diverse time series domains.
・Insofar as scaling has been a critical driver of performance of foundation models in other modalities such as language and vision, much recent work on time series foundation modeling has focused on scaling.
cs.LG updates on arXiv.org
Risk reversal for least squares estimators under nested convex constraints
・arXiv:2601.16041v2 Announce Type: replace-cross Abstract: In constrained stochastic optimization, one expects that restricting the feasible set, provided it still contains the true parameter, should not increase the statistical risk of the corresponding projection estimator.
・We show that this intuition can fail, even in basic settings.
・We investigate this phenomenon in the Gaussian sequence model.
cs.LG updates on arXiv.org
Robustness and Cybersecurity in the EU Artificial Intelligence Act
・arXiv:2502.16184v3 Announce Type: replace-cross Abstract: The EU Artificial Intelligence Act (AIA) establishes different legal principles for different types of AI systems.
・While prior work has sought to clarify some of these principles, little attention has been paid to robustness and cybersecurity.
・This paper aims to fill this gap.
cs.LG updates on arXiv.org
Sampling Decisions: Exact Path-Space Correction, Prior Cancellation and Local-Boltzmann Guidance
・arXiv:2503.14549v3 Announce Type: replace Abstract: How can a cheap but biased sequential, finite-horizon sampler over a discrete space be corrected so that its terminal output follows a prescribed Gibbs distribution?
・We formulate Sampling Decisions as a path-space relative-entropy projection on a growing autoregressive state graph.
・The unique prior-relative minimizer is a Doob transform governed by a linear backward
The Verge
Samsung’s Galaxy Z Fold 8 feels like the future
・A short king.
・Is this the future of foldable phones?
・Samsung clearly thinks so.
Zennの「機械学習」のフィード
SARSA・Q学習・方策勾配法をゼロから整理する
・この記事のゴール 強化学習を勉強していると、こんな疑問にぶつかります。
・SARSA と Q学習ってほとんど同じ式なのに、何が違うの？どう使い分けるの？ 2つを組み合わせることはできないの？ そもそもこれらは「何を最大化」しているの？状態の価値？行動の価値？ 方策勾配法（Policy Gradient）はまた別の話なの？ 「Model-Free / Model-Based」という分け方は、価値ベース/方策ベースと何が違うの？ この記事では、これらを 1枚の地図 の上に並べて整理します。数式は出てきますが、初学者でもイメージできるように「気持ち」から説明します。
stat.ML updates on arXiv.org
SCOPE-FE: Structured Control of Operator and Pairwise Exploration for Feature Engineering via Quality-Aware Candidate-Space Reduction
・arXiv:2604.27025v2 Announce Type: replace Abstract: Automatic feature engineering can improve predictive performance on tabular data by generating diverse feature transformations.
・However, the candidate space induced by combinations of input features and operators grows rapidly with dimensionality, resulting in substantial computational cost.
・We propose SCOPE-FE, a framework that controls the search space before cand
cs.LG updates on arXiv.org
Seesaw: Accelerating Training by Balancing Learning Rate and Batch Size Scheduling
・arXiv:2510.14717v2 Announce Type: replace Abstract: Increasing the batch size during training -- a ''batch ramp'' -- is a promising strategy to accelerate large language model pretraining.
・While for SGD, doubling the batch size can be equivalent to halving the learning rate, the optimal strategy for adaptive optimizers like Adam is less clear.
・As a result, any batch-ramp scheduling, if used at all, is typically tuned
cs.LG updates on arXiv.org
Self-Attention Dynamics with Rotary Position Embeddings: Twisted States and Explicit Consensus Rates on the Sphere
・arXiv:2607.24502v1 Announce Type: cross Abstract: Rotary position embeddings (RoPE) modify attention scores through position-dependent rotations, but their effect on normalized token dynamics is not captured by the vanilla spherical self-attention model.
・We study the continuous-time dynamics obtained when queries and keys are rotated while values remain on the unit sphere.
・The resulting attention kernel is reversible
cs.LG updates on arXiv.org
Self-Distillation of Hidden Layers for Self-Supervised Representation Learning
・arXiv:2603.15553v2 Announce Type: replace-cross Abstract: The landscape of self-supervised learning (SSL) is currently dominated by generative approaches (e.g.
・MAE) that reconstruct raw low-level data, and predictive approaches (e.g.
・I-JEPA) that predict high-level abstract embeddings.
cs.LG updates on arXiv.org
Self-Motivated Growing Neural Network for Adaptive Architecture via Local Structural Plasticity
・arXiv:2512.12713v2 Announce Type: replace-cross Abstract: Control policies are often implemented with fixed-capacity multilayer perceptrons trained by backpropagation, which require architecture selection in advance and cannot adapt their capacity during learning.
・This paper introduces the Self-Motivated Growing Neural Network (SMGrNN), a gradient-trained controller whose topology evolves online through a local Struc
stat.ML updates on arXiv.org
Sequential Preconditioned Conjugate Gradient Method for Linear Statistical Models
・arXiv:2607.25272v1 Announce Type: cross Abstract: We propose a randomized iterative method for the ordinary least-squares estimation problem in large-scale linear statistical models, namely the Sequential Preconditioned Conjugate Gradient Method (SPCG).
・SPCG constructs a sequence of sketched least-squares subproblems with increasing sketch sizes, applies PCG as the inner solver, and warm-starts each subproblem from t
stat.ML updates on arXiv.org
Sharpness-Aware Minimization and Muon: Robustness under the Spectral Norm
・arXiv:2607.26001v1 Announce Type: cross Abstract: Sharpness-Aware Minimization (SAM) aims to improve generalization by encouraging insensitivity to small, worst-case parameter perturbations.
・However, the notion of a "small" perturbation is inherently geometry-dependent: while existing SAM variants have explored a wide range of choices, a clear perspective on which geometries are most effective in practice remains elu
cs.LG updates on arXiv.org
Sheaf-Laplacian Obstruction and Projection Hardness for Cross-Modal Compatibility on a Modality-Independent Site
・arXiv:2604.07632v2 Announce Type: replace Abstract: Cross-modal representations vary in how easily they can be aligned, and compatibility is generally non-transitive: two modalities may align through an intermediate modality at lower complexity than through a direct map.
・We introduce a reference formalism that evaluates all modalities on a fixed neighborhood site and defines two directed invariants.
・Projection hardne
cs.LG updates on arXiv.org
Shift-Aware Calibration for Fine-Tuned CLIP: Leveraging Image-Text Alignment
・arXiv:2501.19060v4 Announce Type: replace-cross Abstract: Vision-language models (VLMs), such as CLIP, adapt effectively to downstream tasks through prompt tuning, but fine-tuning can misalign predictive confidence and accuracy, particularly on unseen classes.
・Existing VLM-specific calibration methods mainly rely on textual features of train classes, limiting their applicability across class and distribution shifts.
cs.LG updates on arXiv.org
Sign-Symmetry Learning Rules are Robust Fine-Tuners
・arXiv:2502.05925v2 Announce Type: replace Abstract: Backpropagation (BP) has long been the predominant method for training neural networks due to its effectiveness.
・However, numerous alternative approaches, broadly categorized under feedback alignment, have been proposed, many of which are motivated by the search for biologically plausible learning mechanisms.
・Despite their theoretical appeal, these methods have cons
cs.LG updates on arXiv.org
SimBEV2X: A Large-Scale Dataset and Data Generation Tool for Multi-Task Vehicle-to-Everything Cooperative Perception
・arXiv:2607.23910v1 Announce Type: cross Abstract: Cooperative perception through vehicle-to-everything (V2X) communication can overcome the inherent physical limitations of individual autonomous vehicles, such as occlusions and limited sensor range.
・However, the development of robust V2X algorithms, particularly those relying on unified spatial representations like bird's-eye view (BEV) representation, is hampered by
cs.LG updates on arXiv.org
Smooth Learning with Hard Constraints via Legendre-Regularized Policies
・arXiv:2607.24007v1 Announce Type: cross Abstract: We revisit contextual optimization from the perspective of policy class design.
・A desirable policy class should be expressive enough to learn rich context-decision relationships, should enforce hard feasibility constraints rather than soft penalty terms, and should remain smooth enough for gradient-based training on downstream decision losses.
・Existing approaches usua
cs.LG updates on arXiv.org
SpecBox: Speculative Sandbox Scheduling for Efficient LLM Agent Serving
・arXiv:2607.23933v1 Announce Type: cross Abstract: As LLM agents increasingly rely on the Model Context Protocol (MCP) to invoke isolated external sandboxes, disaggregated sandbox deployment introduces a fundamental tension between resource utilization and interactive tail latency.
・Persistent long-lived sandbox reservations incur excessive memory overhead at scale, while lazy on-demand instantiation generates severe c
cs.LG updates on arXiv.org
SpecFormer: Mitigating Embedding and Attention Collapse via Spectral-Aware Transformer for Recommendation
・arXiv:2607.24025v1 Announce Type: cross Abstract: Transformer architectures have achieved remarkable success across diverse domains; however, directly applying their standard self-attention mechanism to recommendation often yields suboptimal performance, sometimes even trailing behind well-designed simple recommendation models.
・In this paper, we reveal that this performance bottleneck stems from severe embedding and
cs.LG updates on arXiv.org
Stacking the Deck: Tunable Trainability in Stacked LCUs
・arXiv:2607.24686v1 Announce Type: cross Abstract: Variational quantum circuits have been central to many proposed near-term applications of quantum computing, but a growing body of evidence suggests that trainability and quantum advantage are fundamentally at odds: ans\"atze expressive enough to resist efficient classical simulation tend to exhibit barren plateaus, while structures that provably rule out barren plate
stat.ML updates on arXiv.org
Standard Transformers Achieve the Minimax Rate in Nonparametric Regression with $C^{s,\lambda}$ Targets
・arXiv:2602.20555v2 Announce Type: replace Abstract: The tremendous success of Transformer models in fields such as large language models and computer vision necessitates a rigorous theoretical investigation.
・To the best of our knowledge, this paper is the first work proving that standard Transformers can approximate H\"older functions $ C^{s,\lambda}\left([0,1]^{d\times n}\right) $$ (s\in\mathbb{N}_{\geq0},0<\lambda\
cs.LG updates on arXiv.org
Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis
・arXiv:2605.03160v2 Announce Type: replace Abstract: The standard protocol for interpreting sparse-autoencoder (SAE) features labels each feature from its top-activating contexts and validates the label by steering that single feature at a typical magnitude.
・We argue that this inspects one cell of a larger steering grid, steering condition (single feature, joint feature set, matched random direction) crossed with stee
cs.LG updates on arXiv.org
Stochastic Counterdiabatic Driving via Biorthogonal Liouvillian Eigenmodes
・arXiv:2607.24393v1 Announce Type: cross Abstract: Finite-time driving of stochastic systems generates excess dissipation, causing the evolving probability distribution to lag behind the instantaneous equilibrium, and consequently degrading the convergence of nonequilibrium free energy estimators based on the Jarzynski equality.
・Escorted free energy simulations address the non-adiabatic lag by engineering control fiel
Zennの「大規模言語モデル」のフィード
TAKTのモデル構成、高難易度タスクでも「動かすと悪化」は本当か？issue #23で追試
・はじめに 前回の記事では、AIオーケストレーションツールTAKTで実装（codingタグ）・レビュー（reviewタグ）に割り当てるモデルの強度を変えると、中難易度issueにおいてはほぼ例外なくコストが悪化するという結果が得られました。悪化幅には非対称性があり、「レビューを弱める」方向だけがほぼ無風（+4%）で、それ以外（実装を強める/弱める、レビューを強める）は+45〜81%の悪化でした。
・ただし前回の記事では、この結論に対して次のような限界を挙げていました。
・上位モデルの推論能力が真に効いてくるような、設計判断や複雑な依存関係の把握が必要な高難易度タスクにおいても同じ結論が成...
stat.ML updates on arXiv.org
TaylorPODA: A Taylor Expansion-Based Method to Improve Post-Hoc Attributions for Opaque Models
・arXiv:2507.10643v4 Announce Type: replace Abstract: Post-hoc model-agnostic local attribution (LA) methods have been widely adopted to explain opaque AI models by quantifying feature-wise contributions.
・However, many existing methods rely on heuristic or only partially justified attribution mechanisms, while the quality of attribution itself is often shaped by downstream objectives without universally accepted standa
cs.LG updates on arXiv.org
TEmBed-T: A Multi-Dimensional Benchmark for Table-Level Embeddings
・arXiv:2607.24130v1 Announce Type: cross Abstract: Tabular data is the dominant structured-data modality, and learning table representations has become a core research direction.
・Table-level embeddings in particular underpin a wide range of applications, including table retrieval, data lake discovery, and table classification.
・Despite their importance, there is still limited understanding of how different embedding ap
Hugging Face Papers
Temporal-Distance JEPA: Plan-Aware Representation Learning for Latent World Model Predictive Control
Temporal-Distance JEPA: Plan-Aware Representation Learning for Latent World Model Predictive Control
WIRED
The 15 Best Pool Accessories to Upgrade Your Summer (2026)
・These are the cleaning robots, water monitors, and toys actually worth buying for pool season.
cs.LG updates on arXiv.org
The balance between compactness and forecast accuracy of data-driven latent-space reduced-order models in controlled wake flows
・arXiv:2607.24569v1 Announce Type: cross Abstract: Model-based active flow control requires predictive models that are accurate, stable, and fast enough for real-time optimisation.
・In controlled wake flows, this is often achieved through Reduced-Order Models (ROMs) that first compress high-dimensional velocity snapshots into a latent space and then learn a time- stepping predictor for the dynamics in the latent space.
stat.ML updates on arXiv.org
The Barron-Lipschitz Energy Gap and Depth Separation Phenomena in Scientific Machine Learning
・arXiv:2607.25905v1 Announce Type: cross Abstract: We illustrate in several examples that even neural networks of infinite width (specifically, Barron functions) may encounter substantial obstacles when used as a model class for problems in the calculus of variations.
・An instance of practical relevance concerns the bending, stretching and folding of a thin elastic shell with anchored or clamped boundary conditions whe
The Verge
The Ferrari Luce has at least 500 fans
・Just two months after a bumpy launch earlier this year, Ferrari has reportedly already hit its 2026 sales goal for its polarizing, Jony Ive-designed Luce EV.
・Ahead of Ferrari announcing its quarterly earnings on Thursday, the Financial Times reports that Ferrari was aiming to sell "just under 500" Luce units this year, and reached that target this month.
・Luce sales so far were reportedly driven by "strong demand from
cs.LG updates on arXiv.org
The Hitchhiker's Guide to Agentic AI: From Foundations to Systems
・arXiv:2606.24937v2 Announce Type: replace-cross Abstract: The Hitchhiker's Guide to Agentic AI is a comprehensive practitioner's reference for building autonomous AI systems.
・The book covers the full stack from first principles to production deployment, organized around a central thesis: building great agentic systems requires understanding every layer of the pipeline, not just one.
・The book opens with the LLM substr
cs.LG updates on arXiv.org
The Illusion of Secure LLM Code: Closing the Security Gap via Iterative Reprompting
・arXiv:2607.23710v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly integrated into software development workflows, yet their ability to autonomously generate secure authentication code remains uncertain.
・This paper evaluates the security architecture of authentication systems generated by five prominent AI coding assistants through a bi-modal assessment framework combining static code ana
cs.LG updates on arXiv.org
The Interplay Between Interpolation and Aggregation in Regression: Optimal Sample Complexity
・arXiv:2605.29819v2 Announce Type: replace Abstract: This work investigates theoretically the interplay between interpolation and aggregation in regression.
・We establish that the $\gamma$-graph dimension characterizes learnability for a broad class of natural aggregation procedures.
・Furthermore, we prove that an extremely simple aggregation procedure, combining three interpolating hypotheses via the median, is optimal
The Verge
The Nothing Ear 3A look great… and sound good enough
・Nothing has had a strong visual identity since the Ear 1 were released in 2021 for $99 and challenged the notion that a highly featured headset requires a high price.
・The Ear 3A have the same transparent look as Nothing's previous earbuds, with a clear case and stem design accented by a solid-colored body - this time adding a striking pink alongside yellow, black, and white options.
・And Nothing continues to improve b
cs.LG updates on arXiv.org
The Optimal Sample Complexity of Linear Contracts
・arXiv:2601.01496v3 Announce Type: replace-cross Abstract: In this paper, we settle the problem of learning optimal linear contracts from data in the offline setting, where agent types are drawn from an unknown distribution and the principal's goal is to design a contract that maximizes her expected utility.
・Specifically, our analysis shows that the simple Empirical Utility Maximization (EUM) algorithm yields an $\var
cs.LG updates on arXiv.org
The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
・arXiv:2607.24720v1 Announce Type: cross Abstract: Multi-turn long-horizon planning is critical for foundation model agents, yet how to fundamentally improve it remains unclear.
・Existing models are trained on uncontrollable and opaque Internet data, making it difficult to identify how planning ability is acquired, shaped, and integrated.
・To address this challenge, we introduce a unified and controlled multi-turn envir
stat.ML updates on arXiv.org
The Ray Tracing Sampler: Bayesian Sampling of Neural Networks for Everyone
・arXiv:2510.25824v2 Announce Type: replace-cross Abstract: We derive a family of Markov Chain Monte Carlo (MCMC) sampling methods based on following ray paths in a medium where the refractive index $n(x)$ is a function of the desired likelihood $\mathcal{L}(x)$, extending past work on isokinetic sampling.
・The simplest ray tracing method propagates rays at constant speed through parameter space, leading to orders of ma
cs.LG updates on arXiv.org
The Tokenizer Tax: Quantifying and Explaining the Cross-Lingual Cost of Subword Tokenization for Indian Languages
・arXiv:2607.24276v1 Announce Type: cross Abstract: Large language models (LLMs) process text through subword tokenizers rather than directly reading characters or words.
・Because these tokenizers are trained predominantly on English-centric corpora, they introduce a systematic and often overlooked disadvantage for many non-English languages.
・In this work, we quantify this tokenizer tax for Indian languages using the FL
cs.LG updates on arXiv.org
The Zero Pattern of a Design Matrix Drives Multiple Descent in Over-parameterized Regression
・arXiv:2607.24041v1 Announce Type: cross Abstract: Over-parameterized linear regression has been widely studied over the last decade.
・However, most existing works assume that the covariates are independent and that their covariance matrices are non-degenerate.
・In this paper, we relax both assumptions and derive deterministic equivalents for the prediction risk in a vanishing-ridge regime.
WIRED
Therabody Promo Codes: 15% Off July 2026
・Save on the science-backed devices you’ve been eyeing with 15% off Theragun discount code and 30% off other great deals.
cs.LG updates on arXiv.org
To Erase, or Not to Erase: Robust Training-Free Concept Erasure with Preservation aware Adaptive Ranked Subspace Expansion
・arXiv:2607.23492v1 Announce Type: cross Abstract: Concept erasure techniques (CETs) edit text-to-image diffusion models to erase undesired targets such as NSFW content or copyrighted styles, while preserving model utility on benign concepts.
・Current CETs face a trade-off between erasure robustness and utility: stronger edits erase the target more reliably but degrade utility on non-target concepts, and vice versa.
cs.LG updates on arXiv.org
TopoFE: topology-aware LLM-guided Automated Feature Engineering
・arXiv:2607.23286v1 Announce Type: cross Abstract: Automatic feature engineering (AutoFE) for tabular learning can be naturally formulated as a program synthesis problem, where the objective is to discover predictive feature transformations from an exponentially large search space.
・Recent advances in large language models (LLMs) have expanded the expressiveness of AutoFE by enabling feature program generation beyond p
stat.ML updates on arXiv.org
Towards Efficient Inference under Nonmonotone Missingness with General Imputation
・arXiv:2509.24158v2 Announce Type: replace-cross Abstract: Missing data are ubiquitous in classical survey and longitudinal studies as well as modern multi-modality data analysis.
・A longstanding challenge arises under nonmonotone missingness, where different units may observe arbitrary subsets of all variables.
・We study parameter estimation and inference problem under this setting.
Hugging Face Papers
Towards Robust Reinforcement Learning for Small-Scale Language Model Agents
Towards Robust Reinforcement Learning for Small-Scale Language Model Agents
cs.LG updates on arXiv.org
Transfer learning for conflict and duplicate detection in software requirement pairs
・arXiv:2301.03709v3 Announce Type: replace-cross Abstract: Consistent and holistic expression of software requirements is important for the success of software projects.
・In this study, we aim to enhance the efficiency of the software development processes by automatically identifying conflicting and duplicate software requirement specifications.
・We formulate the conflict and duplicate detection problem as a requiremen
stat.ML updates on arXiv.org
Transfer Learning in High-Dimensional Clustering: Minimax Thresholds and Applications in Single-Cell Data
・arXiv:2607.25031v1 Announce Type: cross Abstract: Clustering is a fundamental problem in statistics, with applications across many scientific disciplines.
・In many modern applications involving clustering, the primary dataset (the target data) is accompanied by related datasets (the source data).
・Transferring information from such sources may improve clustering accuracy in the target, making transfer learning for clus
cs.LG updates on arXiv.org
TRIDENT: Benchmarking LLM Safety in Finance, Medicine, and Law
・arXiv:2507.21134v2 Announce Type: replace-cross Abstract: As large language models (LLMs) are increasingly deployed in high-risk domains such as law, finance, and medicine, systematically evaluating their domain-specific safety and compliance becomes critical.
・While prior work has largely focused on improving LLM performance in these domains, it has often neglected the evaluation of domain-specific safety risks.
cs.LG updates on arXiv.org
TriShieldRAG: A Three-Ring Defense-in-Depth Framework Against Knowledge Corruption in Retrieval-Augmented Generation
・arXiv:2607.23838v1 Announce Type: cross Abstract: Retrieval-Augmented Generation (RAG) lets a large language model answer questions using documents retrieved from an external knowledge base at query time.
・This makes RAG useful for private data, fast-changing information, and reducing hallucination, but it also means the model's answer is only as trustworthy as whatever the retriever hands it.
・If the knowledge base ac
WIRED
Tropical Diseases Like Dengue Fever and Chikungunya Are on the Rise in Europe
・Climate change is helping invasive species to not only establish themselves across the continent but to stay alive for longer.
・Scientists suggest we get used to it.
cs.LG updates on arXiv.org
TRUAV: Distributed Multi-Agent Reinforcement Learning for Trajectory Planning and Routing Enhancement in UAV-Aided IoT-Enabled VANETs
・arXiv:2607.23734v1 Announce Type: cross Abstract: Unmanned aerial vehicles (UAVs) have emerged as a key enabler of next-generation Internet of Things (IoT) ecosystems, offering flexible aerial relaying to extend connectivity across dynamic vehicular ad hoc networks (VANETs) in smart city environments.
・However, conventional centralized approaches for UAV trajectory planning require continuous global network state aggr
cs.LG updates on arXiv.org
TSP with Predictions: Heatmap to Tour with Provable Guarantees
・arXiv:2607.03791v2 Announce Type: replace-cross Abstract: The Traveling Salesperson Problem (TSP) has long served as a benchmark for evaluating the strength of optimization techniques in the classical theory of algorithms.
・In recent efforts to apply ML to algorithmic problems, TSP has also become a natural testbed for the development of ML-based techniques.
・A common approach is to train a neural network to output a h
cs.LG updates on arXiv.org
Two-Timescale Hierarchical Reinforcement Learning for Resilient Operations
・arXiv:2607.23434v1 Announce Type: cross Abstract: Unexpected shocks recur in global operations, requiring decision rules that adapt as market and operating conditions change.
・Many operational systems also have hierarchical structures in which long-term and short-term decisions pursue a shared objective.
・We study how hierarchical reinforcement learning can strengthen resilience by adapting these interdependent rules j
cs.LG updates on arXiv.org
Uncertainty Modeling for Multi-Objective RTA Interception with Distillation Acceleration
・arXiv:2511.05582v3 Announce Type: replace Abstract: Real-Time Auction (RTA) interception decides which incoming advertising requests reach downstream systems, and therefore controls the quality of the data those systems learn from.
・At JD.com, off-site advertising produces on the order of hundreds of billions of requests per day, and the RTA channel alone serves up to hundreds of millions of requests per minute.
stat.ML updates on arXiv.org
Unifying Active Learning and Semi-Supervised Learning for Medical Image Segmentation
・arXiv:2607.25014v1 Announce Type: cross Abstract: In practical settings, medical image segmentation models are often developed with limited annotated data rather than fully labeled datasets.
・Training frequently begins in ultra-low labeled regimes where only a small number of volumes are annotated.
・In such scenarios, practitioners must simultaneously decide which cases to annotate and how to best use the remaining unl
cs.LG updates on arXiv.org
Universal Decision Learners
・arXiv:2605.30694v2 Announce Type: replace Abstract: Many theories of decision making -- planning, reinforcement learning, causal intervention, online learning, and game-theoretic equilibrium -- turn local information into globally coherent behavior.
・This paper proposes a common categorical formulation: a Universal Decision Learner (UDL) extends a partially specified decision functor from observed contexts to new cont
cs.LG updates on arXiv.org
Unraveling the Mechanism of Drug Binding to SARS-CoV-2 RNA Pseudoknot with Thermodynamics-Driven Machine Learning
・arXiv:2604.14906v4 Announce Type: replace-cross Abstract: The pseudoknot secondary structure in SARS-CoV-2 RNA is essential for regulating protein synthesis through $-$1 programmed ribosomal frameshifting ($-1$ PRF), a mechanism that allows the virus to generate both structural and non-structural proteins from overlapping reading frames.
・This pseudoknot exhibits both threaded and unthreaded long-lived topologies.
cs.LG updates on arXiv.org
Using Reinforcement Learning to Optimize the Global and Local Crossing Number
・arXiv:2509.06108v3 Announce Type: replace-cross Abstract: Graph drawing concerns the algorithmic visualization of graphs.
・A good drawing of a graph is easy to read and facilitates solving tasks on the graph.
・Several properties have been identified to occur in good drawings of graphs.
cs.LG updates on arXiv.org
Variational Neural Belief Parameterizations for Robust Dexterous Grasping under Multimodal Uncertainty
・arXiv:2604.25897v2 Announce Type: replace-cross Abstract: Contact variability, sensing uncertainty, and external disturbances make grasp execution stochastic.
・Expected-quality objectives ignore tail outcomes and often select grasps that fail under adverse contact realizations.
・Risk-sensitive POMDPs address this failure mode, but many use particle-filter beliefs that scale poorly, obstruct gradient-based optimization,
cs.LG updates on arXiv.org
Variational Quantum Conditional Boltzmann Machines for Time-Series Forecasting: Architectures, Symmetric Hyperparameter Evaluation, and a Nonlinear Benchmark
・arXiv:2607.24065v1 Announce Type: cross Abstract: In this study, we developed and evaluated four conditional energy-based forecasting architectures: a classical Gaussian-Bernoulli CRBM, a hybrid quantum-classical QCRBM, a full-register QQRBM, and a lag-feature QFeatureQRBM with complete derivations of their conditional distributions, Contrastive-Divergence gradients, and hybrid training, bridging the energy-based for
stat.ML updates on arXiv.org
Vector-Valued Distributional Reinforcement Learning Policy Evaluation: A Hilbert Space Embedding Approach
・arXiv:2601.18952v2 Announce Type: replace-cross Abstract: We propose an (offline) multi-dimensional distributional reinforcement learning framework (KE-DRL) that leverages Hilbert space mappings to estimate the kernel mean embedding of the multi-dimensional value distribution under a proposed target policy.
・In our setting, the state-action variables are multi-dimensional and continuous.
・By mapping probability measure
Zennの「大規模言語モデル」のフィード
Vibe Quant Insight #001 — オープンウェイトの逆襲、アップルの逆説、そして自ら侵入したAI
・Vibe Quant Insight #001 2026年7月29日 · 創刊号 뉴스レター · News Letter · 🇯🇵 日本語 AI・クオンツ・セキュリティが交差する地点を、週に一通。ニュースではなく、判断のための材料をお届けします。
・各言語版について — 本ニュースレターで扱うコラムの英語版・日本語版は、docs.vibequant.cc/columns/ の各詳細リンクからご確認いただけます。主要コラムの多くは韓国語・英語・日本語・中国語の4言語で発行しています。
・今号の要約 セクション テーマ 一行で 1.
Hugging Face Papers
Visual prompt engineering for video models
Visual prompt engineering for video models
Hugging Face Papers
VisualPatchWorld: Code World Models as Latent Structured Representations for Planning
VisualPatchWorld: Code World Models as Latent Structured Representations for Planning
cs.LG updates on arXiv.org
VLASH: Real-Time VLAs via Future-State-Aware Asynchronous Inference
・arXiv:2512.01031v2 Announce Type: replace-cross Abstract: Vision-Language-Action models (VLAs) are becoming increasingly capable across diverse robotic tasks.
・However, these models are typically deployed under synchronous inference, where the robot waits for model inference to complete before acting, and cannot perceive or respond to environmental changes during action execution.
・This not only introduces noticeable a
Google DeepMind News
We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
The Verge
We’re running out of reasons to ignore AI safety
・Earlier this month, OpenAI gave several of its AI models a task: complete a test designed to measure their cybersecurity capabilities.
・It put the systems in a sandboxed environment without an internet connection and set them off to work.
・What happened next is almost laughably silly - but also, as Adam Gleave, cofounder and CEO of AI safety organization FAR.AI, put it, "a visceral example of how misaligned AI could ca
cs.LG updates on arXiv.org
WeCon: An Efficient Weight-Conditioned Neural Solver for Multi-Objective Combinatorial Optimization Problems
・arXiv:2605.22876v2 Announce Type: replace Abstract: Existing neural solvers for Multi-Objective Combinatorial Optimization Problems (MOCOPs) commonly adopt decomposition-based strategies that scalarize a MOCOP into multiple subproblems associated with distinct weight vectors.
・However, they either inject weights only once during decoding, limiting weight-conditioned context modeling, or primarily during encoding, caus
cs.LG updates on arXiv.org
Weighted Low-Rank Matrix Approximation: Acceleration and Applications
・arXiv:2109.11057v2 Announce Type: replace-cross Abstract: Weighted low-rank matrix approximation (WLRMA) generalizes classical low-rank approximation and matrix completion by allowing arbitrary elementwise weights.
・Such formulations arise naturally in a broad class of statistical models, including generalized linear low-rank models, where WLRMA serves as the computational primitive for parameter estimation.
cs.LG updates on arXiv.org
What You See Is What You Get: Observation-Aligned Supervision for Chart-to-Code Generation
・arXiv:2607.04726v2 Announce Type: replace-cross Abstract: Chart-to-code generation is commonly trained with supervised fine-tuning on reference plotting scripts, implicitly treating the gold code as a fully observable target.
・We argue that this assumption is often invalid: many chart programs contain latent raw variables that cannot be uniquely recovered from the rendered image.
・We identify this systematic latent--ob
cs.LG updates on arXiv.org
When Average Calibration Fails: Site-Conditional Federated Conformal Risk Control
・arXiv:2606.20115v3 Announce Type: replace Abstract: Conformal risk control (CRC) provides distribution-free segmentation guarantees by calibrating a prediction-set threshold on held-out data.
・In federated deployments, the standard approach pools calibration scores into a single threshold.
・We quantify, on real multi-institutional brain tumor data (FeTS-2022, 1,251 subjects, 20 institutions), a critical failure: naive
cs.LG updates on arXiv.org
When Every Simulation Counts: Value-Based Reinforcement Learning for Accelerated Photonics Inverse Design
・arXiv:2607.23469v1 Announce Type: cross Abstract: Photonic-crystal surface-emitting lasers (PCSELs) can combine high-power operation with narrow-divergence surface emission, but optimizing coupled parameters requires costly full-wave simulations.
・Deep Q-network (DQN) optimization can reuse simulated transitions to guide edits, yet which value-learning mechanisms remain reliable under tight simulation budgets is unkno
cs.LG updates on arXiv.org
When Good Equations Get Bad Scores: Improving Symbolic Regression Through Better Parameter Optimization
・arXiv:2605.23272v3 Announce Type: replace Abstract: Symbolic Regression (SR) plays a central role in scientific knowledge discovery by distilling mathematical equations from observational data.
・Most existing SR methods function within a bi-level optimization framework: an outer loop that searches for the discrete equation structure, and an inner loop that optimizes the continuous parameters of that structure.
cs.LG updates on arXiv.org
When LLM Defenses Backfire: Characterizing Safety, Performance, and Cost Trade-offs
・arXiv:2607.24392v1 Announce Type: cross Abstract: Jailbreak defenses are essential for protecting large language models (LLMs), but they can also introduce secondary costs that weaken model utility.
・We present a systematic study of these defense trade-offs along three dimensions: performance impact, over-refusal on benign inputs, and inference cost.
・Rather than treating defenses as a single class, we organize them by
cs.LG updates on arXiv.org
When Low CER is Not Enough: An Analysis of Hallucinations in Vision-Language OCR Systems on Historical Uruguayan Documents
・arXiv:2607.24077v1 Announce Type: cross Abstract: Optical Character Recognition (OCR) is a key component in the digitization of historical archives.
・Recently, Vision-Language Models (VLMs) have emerged as strong alternatives to traditional OCR systems, achieving state-of-the-art performance on standard benchmarks.
・However, their suitability for archival transcription remains insufficiently understood.
cs.LG updates on arXiv.org
When Rates Are Geometric: Rate-Certificate Transfer for Contact Splittings in Optimization
・arXiv:2607.23642v1 Announce Type: cross Abstract: Discrete optimization algorithms are often analyzed through continuous-time limiting ODEs, but a convergence certificate for the ODE is not automatically one for the discrete algorithm.
・We develop contact Hamiltonian systems as a setting where the transfer can be made precise.
・A contact Hamiltonian $H$ on $J^1(\mathbb{R}^n)$ obeys the intrinsic decay identity $\dot H
Hugging Face Papers
Where Quality Breaks in Compressed Short-Text Generation: Staged Bottleneck Localization
Where Quality Breaks in Compressed Short-Text Generation: Staged Bottleneck Localization
cs.LG updates on arXiv.org
Where Quality Breaks in Compressed Short-Text Generation: Staged Bottleneck Localization
・arXiv:2607.24176v1 Announce Type: cross Abstract: Compressed short-text generators can fail in two different places: the codec may discard information before generation starts, or the latent generator may produce weak codes.
・Without separating these failure modes, researchers can spend compute improving the wrong component.
・We study this problem in a controlled 64-to-16 TinyStories case study built from a hierarchica
WIRED
X Says Australia’s Under-16 Social Media Ban Risks Interfering With Foreign Law
・Elon Musk’s platform attacked the country’s “invasive” information-gathering powers as it demands the government drop efforts to strengthen its no-children policy.
The Verge
Xbox outage shouldn’t have affected games on disc, Microsoft confirms
・After the recent Xbox outage even blocked people from playing offline disc-based console games, Microsoft is trying to soothe concerns that you don't truly own anything under Xbox's license system.
・In a statement to The Verge, Xbox technology chief Scott Van Vliet says that while physical game discs are subject to licensing inspections, that shouldn't have prevented them from working during this week's outage, or whi
cs.LG updates on arXiv.org
XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control
・arXiv:2607.04171v2 Announce Type: replace-cross Abstract: Large Vision-Language Models (LVLMs) have shown strong multimodal understanding and spatial grounding, but their computational cost limits real-time robotic control.
・In contrast, lightweight models are suitable for edge deployment but often suffer from "spatial blindness", namely weak native spatial prediction ability.
・Training Vision-Language-Action (VLA) mod
ITmedia NEWS 最新記事一覧
イオンモール熊本内部の様子も──自衛隊や警察庁、救助活動の動画・写真をSNSで積極発信
・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、警察庁や自衛隊などが救助活動を進めている。XなどのSNSでも活動の様子を積極的に発信しており、画像や映像を通して動向を確認できる。
#LLMタグ
これはOCRではない——NVIDIAの新モデル「Llama Nemotron Rerank VL」が画像PDFにもたらす価値
・NVIDIAがHugging Faceで公開した「llama-nemotron-rerank-vl-1b-v2-fp8」。
LLMタグが付けられた新着記事 - Qiita
ストーリーポイントは「人間がやる当てずっぽう」だった。アジャイル開発の一次見積もりはAIの仕事にする
・この記事の結論 ストーリーポイントは客観的な測定値ではなく、不完全な情報とチームの経験から導く主観的な相対推定 過去事例の検索、相対比較、仮採点、根拠整理、不確実性の検出はAIへ移管できる 人間は、AIが取得できない文脈を補い、認識差を解消し、最終的なサイジングに責...
#LLMタグ
ツールを使える人より、役割を設計できる人が生き残る——AIエージェント時代の職種再定義
・「AIツールを使いこなせる人材が求められる」 —— この言説、そろそろ賞味期限切れです。2024年から2025年にかけて、MITのAI Agent Indexによれば追跡対象30エージェントのうち24がリリースまたは更新されました。自律性のレベルは急速に上がっています。問われているのは「ツールを使えるか」ではなく、「エージェントの設計と運用のどこに自分が立つか」です。技術実装ができる人、エージェントの動作を修正できる人、ビジネス成果に焦点を当てて協働する人 —— ZDNETの整理では、この3タイプの人材が今後の組織に不可欠だとされています。
・AIエージェントは今、何を変えようとしているのか 続きをみる
＠IT 全フォーラム 最新記事一覧
ネットワークエンジニア、社内SEって実は「稼げる仕事」？ IT職種別の“単価”に明暗
・ITフリーランスエンジニアの案件・求人サイト「テクフリ」を運営するアイデンティティーが、2026年5月のIT人材市況動向レポートを公表した。IT人材の職種別平均希望単価はどうなっているのか。
Zennの「大規模言語モデル」のフィード
ブラウザ自動化（Playwright）×AIパイプラインのOpenTelemetry導入記
・はじめに 今回の対象は、ブラウザ操作とAI処理を含む自動化ジョブです。
・ログイン → 検索 → ページ内容取得 → AI要約・構造化 → 登録 この一連のパイプラインを、ずっとログだけで運用してきました。
・この記事は、そこにOpenTelemetry(以下OTel)を入れたら運用がどう変わったかの記録です。
Zennの「大規模言語モデル」のフィード
プラグインを足す前に、AIに『会社の規則』を書いた話 ―― 5項目比べて、導入したのは1つだけでした
・こんにちは、AI師範です。
・前回は「『この仕事、誰が拾うの？』――AIセッションを並走させたら、人間の会社と同じ問題が起きた」という話を書きました（「この仕事、誰が拾うの？」）。複数のClaude Codeを並走させると必ず起きる取りこぼしと二重処理を、「待ち番」という当番表の仕組みで外部化した、という話です。
・その記事は、私が普段やっている「Claude Codeを道具ではなく組織として動かす」試みのひとつでした。今回は、その延長線上で、最近気になっていた比較の話をします。テーマはこうです ―― 世間はClaude Codeを「プラグイン」で強化する。私は同じ課題を「制度」で解いていた...
#LLMタグ
マルチエージェント型仮想ディベートアプリを作ってみた
・まあ、作ったのは私ではなくてgeminiなのですが、複数の異なるLLM（AIモデル）が共通の議題に対して「評価・議論・合意形成」を自律的に行うマルチエージェント型ディベートシステムです。
#LLMタグ
モデル性能から実行力へ / 韓国エージェンティックAIイニシアティブ / 日本の第Ⅱ期AI基本計画 雑感
モデル性能から実行力へ / 韓国エージェンティックAIイニシアティブ / 日本の第Ⅱ期AI基本計画 雑感
#AIタグ
運動回数は増えた。でも、買い物には戻れない
・在宅で使えるリハビリアプリが増えています。画面の指示に合わせて体を動かすと、カメラやセンサーがその動きを読み取り、AIが姿勢を採点してくれる。毎日の運動回数も、達成率も、数字ではっきり見えるようになりました。
・でも、その数字が上がることと、「望む生活に戻れること」は、同じでしょうか。
ITmedia NEWS 最新記事一覧
企業が広告を「早く、安全に、大量に」作るAI アドビの新ツールが向かう先は“代理店いらず”か
・生成AIで画像を作るのは速くなった。それでも社外に出す本番の広告には、なかなか使われない。Adobeが7月6日に発表した3つのソリューションは、この「作れるのに使えない」を埋めにきている。ノードをつないで工程そのものを再現可能にする仕組みは、企業とクリエイターのどちらを見ているのか。
ITmedia NEWS 最新記事一覧
熊本「通れた道」マップ、トヨタが公開 ホンダも「通行実績情報マップ」
・28日に熊本県で発生した最大震度7の地震を受け、トヨタ自動車とホンダがそれぞれ、被災エリアで実際に通行できた道を表示するマップを公開した。
ITmedia NEWS 最新記事一覧
熊本地震で「Yahoo!天気」が意外な活躍？ 断水・停電情報など投稿続出 ライフライン情報共有の場に
・熊本県で震度7を観測した地震を巡り、天気アプリ「Yahoo!天気」の投稿機能を使って断水や停電などのライフライン情報を教え合う動きが、XなどのSNSで話題となっている。
ITmedia NEWS 最新記事一覧
警察庁、熊本地震に便乗した詐欺に注意喚起 義援金名目で電子マネーなどだまし取る手口
・警察庁は7月29日、令和8年熊本地震に便乗した詐欺や悪質商法に注意するようXで呼び掛けた。公的機関をかたって義援金を募り、現金や電子マネー、暗号資産をだまし取る手口を挙げている。
ITmedia NEWS 最新記事一覧
災害時、スマホの電池を少しでも長く 携帯各社のバッテリー節電策まとめ
・7月28日に発生した「令和8年熊本地震」の影響で、一部地域では停電や通信障害も発生している。災害時、スマートフォンは家族との連絡や避難情報の確認に欠かせない一方、充電できない状況が長引く可能性もある。NTTドコモ、KDDI、ソフトバンクなどの携帯各社が推奨するスマホのバッテリー消費を抑える方法を整理した。
#AIタグ
作れる。でも、売れるとは限らない。
・ラーメン屋の麺発注ツールが完成した時、僕は本気で思っていました。
・「これは使ってもらえる。」 「店長も絶対楽になる。」 AIでここまで作れる。
・それだけでも十分感動していました。
＠IT 全フォーラム 最新記事一覧
若手ITエンジニアの5割が「管理職になりたくない」 それでも「引き受ける条件」とは？
・キッカケクリエイションは、25～39歳の非管理職ITエンジニアの管理職志望について、「なりたい」「なりたくない」が約5割で拮抗しているとの調査結果を公表した。
Zennの「大規模言語モデル」のフィード
終わったことを、どうやって知る？ 音声からAIエージェントへ仕事を委譲する条件
・音声アシスタントと喋りながら開発を進められたら楽だな、とずっと思っていました。手はキーボードから離れているのに、頭のほうだけ動いている時間ってありますよね。皿を洗っているときとか、散歩中とか。
・なので試してみました。ChatGPT の GPT Live で「こういうものが作りたい」と相談して、その流れのまま実装エージェントへ仕事を渡す。渡す先は Claude Code と Codex の2つです。
・始める前の予想は、正直こうでした。
#AIタグ
中国の無料AIに潜む隠れた代償
・New York TimesのOpinionに、中国の無償AIを使うことによるリスクについて述べてあった。
・The Hidden Cost of China’s Free A.I.
ITmedia NEWS 最新記事一覧
停電中の自宅を離れる時は「ブレーカー落とす」 消費者庁などが「通電火災」の注意喚起
・「災害に伴う停電の復旧後、電気製品の損傷や意図しない作動による火災（通電火災）に注意してください」。消費者庁などが「通電火災」の注意喚起している。
#AIタグ
働くんじゃない。仕組みに働いてもらう。
・AIを触れば触るほど、一つの考えが頭から離れなくなっていました。
・「人が働く時間には限界がある。」 アルバイトをしている時もそうでした。
・頑張れば頑張るほど収入は増える。
＠IT 全フォーラム 最新記事一覧
同志社女子大、「更新のたびに認証システムを止めなきゃならない」をゼロに 何を変えた？
・バージョンアップに伴う認証システムの停止に悩んでいた同志社女子大学。認証システムを見直し、こうした停止を伴う保守作業や、特定の技術者に依存した運用を改善した。どのように実現したのか。