ai Trend Report
Dashboard へ戻る
3
High impact
5
Mid impact
391
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
Adaptive Finite-Budget Training for CVaR Risk-Aware Q-Learning
・arXiv:2608.04305v1 Announce Type: new Abstract: Risk-aware Q-learning (RaQL) provides a model-free, two-timescale estimator for dynamic risk objectives, but its finite-budget behavior remains fragile: fixed inner-loop hyperparameters can produce unstable value estimates, persistent Bellman residuals, and inefficient sample reuse.
・This paper proposes an adaptive training controller for Conditional Value-at-Risk (CVaR)
#AIタグ
AIに仕事を渡せる業務になってなくても、大丈夫。― だからこそ、簡単に作り直せる
・※本記事は個人の見解・考察であり、所属する組織その他の団体を代表するものではありません。また、特定の企業・金融機関・製品・サービスを評価、推奨、批判することを目的としたものではありません。
・業務をデジタル化するとき、 よく言われることがあります。
・「まず現行業務を可視化しましょう」 現行業務をヒアリングする。
The Verge
The left and right agree on one thing: no data centers
・Today, I’m talking with Gaby Del Valle, a policy reporter here at The Verge, about the growing backlash against AI data centers.
・Gaby recently reported a fantastic piece about Hernando County, Florida, where last month the county commission unanimously approved a yearlong moratorium on data center construction.
・She attended a protest there organized by a group called Humans First, a conservative grassroots movement f
#AIタグ
【構造批評】-KOKUSAI韓国DRAM主導編-売上239億円、最大市場は中国から韓国へ移った‼️
・🟧序章｜半導体回復の中心を映した仕向地の逆転 8月6日の日経は、KOKUSAI ELECTRICが2027年3月期の純利益予想を従来の388億円から555億円へ引き上げたと報じました。
・売上高は従来予想から600億円増の3400億円、営業利益は249億円増の794億円です。会社は、生成AI向けの半導体需要を背景に、装置受注が想定以上に強まったと説明しています。
#LLMタグ
AIは人間を裏切ったのか
・2026年8月6日（木）のNACK5『Good Luck!
・Morning!』内「エコノモーニング」では、こんなお話をしました。
・今日は「AIは人間を裏切ったのか」というお話です。
Qiita - 人気の記事
AI僧侶ロボット「ブッダロイド」が見せた、"代行"というフィジカルAIの伸びしろ
・はじめまして。株式会社PRUMでエンジニアをしている、すもも🍑です 日々、プログラミング学習や実務の中で、つまずきやすいポイントや 考え方を整理して発信しています。
・PRUMについて気になった方は、コーポレートサイトもぜひご覧ください。
・▶コーポレートサイト 「AI僧侶...
#AIタグ
ChatGPT無料版と有料版の違い｜2026年版 本当に課金する価値はある？
・ChatGPT無料版と有料版の違い｜どちらを選ぶべき？ 「ChatGPTって無料でも使えるけど、有料版にする意味あるの？」 これまでの記事でChatGPTの使い方やプロンプトを紹介してきましたが、実際に日常的に使い始めると、必ずこの疑問にぶつかります。
・結論から言うと、 - ライトに試したいだけなら無料版で十分 - 仕事で本格的に使うなら有料版のほうが確実に楽 です。
・この記事では、無料版と有料版の違いを「利用制限」「AIの性能」「使える機能」の3つに整理して解説します。
cs.LG updates on arXiv.org
Learning Compression Rules for Network Traffic
・arXiv:2608.04545v1 Announce Type: new Abstract: We study the problem of learning compact rule-based compressors for structured network traffic.
・Each packet is a record of header fields that are highly redundant within a flow, and a compressor is a small set of rules matching such records and replacing predictable fields with short codes.
・We cast rule learning as a two-stage problem: (i) an unsupervised structure-disc
Zennの「大規模言語モデル」のフィード
（LLM Wiki）Karpathy の原典に忠実になってみた
・Karpathy が提唱する LLM Wiki はあくまで概念までしか提示されていないから、「実践したい場合はあなたの環境に合わせて1から構築してね」ということになっている。
・こういうとき、完璧主義の人間としてはオリジナルの思想に忠実でない構築方法（有志によって作成されたテンプレートや、AIが考えたそれらしきルール定義）を許すことができない。
・今回は運よく Claude Code の $1,000 分のサービスクレジットを手に入れることができたので、Fable 5 ならどう考えるか、基本概念に忠実なWikiフォーマットを追求してみた。
Latent.Space
[AINews] Jeff, Sanjay, Oriol, and Quoc depart DeepMind; Demis to Chair; Koray to SVP — what is going on at GDM???
[AINews] Jeff, Sanjay, Oriol, and Quoc depart DeepMind; Demis to Chair; Koray to SVP — what is going on at GDM???
＠IT 全フォーラム 最新記事一覧
「1人1AI」のアプローチは破綻する――チームでAI共有時のセキュリティ問題を解決するベストプラクティス
・Anthropicは、「Claude Tag」における「エージェントアイデンティティー」アクセスモデルの仕組みと、チームのワークスペースでこれを構成する際のベストプラクティスを解説したブログ記事を公開した。
Zennの「大規模言語モデル」のフィード
「AIで生産性10倍」の裏で、英語圏は「2倍」の話を始めている
・情報確認日: 2026-08-06。
・本記事の調査は検索ベースで行い、各情報の確度（複数ソース一致か、単一情報源か）を本文と出典に明記しています。AI関連の数値は鮮度が命です。お読みの時点で状況が変わっている可能性があります。
・日本語のタイムラインでは、いまも「AIで開発生産性10倍」という言葉をよく見かけます。
#LLMタグ
「成功率が落ちた」の正体は、測り方が直っただけだった【測ったつもり編】
・前回は、フォーム営業の自動化で「送信ボタンに辿り着けない」5類型に詰まった話を書きました。
・修正を重ねて、送れるようになった。そう思っていた翌週、送信成功率が急落します。以下は、フォームへの送信を試みた件数の内訳です。
#LLMタグ
【AI速報】Meta、新しいコーディングAIを発表。その横で旧モデルが他社をハックしていたことが判明
【AI速報】Meta、新しいコーディングAIを発表。その横で旧モデルが他社をハックしていたことが判明
#LLMタグ
【FastFlowLM】Ryzen AIのNPUをUbuntuで叩く! 果たして性能は? ～Dockerで環境構築も楽々 Qwen3.6 Gemma4対応～
・こんにちはRcatです。
・今回はとうとうNPUを触っていきます。
・今まで一度も手を付けられていませんでしたが、やっとこの時が来ました。
#LLMタグ
【GPT-Image-2】視力回復‼️左右をよく見て👀✨間違い探し。ChatGPTのOCR解析能力とは🧐
・・今回のテーマ 視力UP間違い無し‼️の間違い探し‼️ ・説明 左側は普通、右側はセクシー 間違いは3箇所。皆さん見つけられるかなぁ👀✨ 続きをみる
#AIタグ
【構造批評】-富士フイルムBI成長空白編-複合機の先に掲げるDX事業は、まだ収益構造が見えない‼️
・🟧序章｜スピンオフが映した複合機の次の課題 8月6日の日経は、富士フイルムホールディングスが、複合機を中核とする富士フイルムビジネスイノベーションのパーシャル・スピンオフを検討し、2〜3年後をめどに上場させる方針を報じました。
・同事業は2026年3月期に売上高1兆1748億円、営業利益637億円を計上し、連結売上高の約35％を占める最大部門です。会社側も、強固な収益基盤と高いキャッシュ創出力を持つ事業と位置づけています。
#LLMタグ
【雑記】5.6 Sol相棒が時報AIになった件
・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。
・最近、5.6 Solが能動的に現在の日時を調べるようになった模様。
・（他のモデルでは未確認） 続きをみる
#LLMタグ
【初心者向け】自動化ツールでThreadsが伸びない3つの原因とLLMで感情を動かす投稿を仕組み化する3ステップ
・「Threadsの運用を自動化すれば、もっと楽に発信を続けられるはずだ」 そう思って自動化ツールを導入したり、ChatGPTに投稿文を書いてもらったりした経験はありませんか？ 続きをみる
#LLMタグ
【生成AIニュース+】『SeedRealtime』『Lyra 2.0』『MiniMax-H3-experimental』『MiniMax-H3-Turbo-Lora』『ComfyUI MiniMax H3 Director』『ComfyUI-Latent-Tiled-PiD』『ComfyUI-MiniMaxH3-Easy』『ComfyUI-DyPE』『MiniMax-H3 Turbo 4-Step』『Muse Spark 1.2 と Muse Code』『Xiaomi-Robotics-1』
・『SuperSplat』 『BirdingPal』 『Human-like AI response concept by Nils Eller』 まいどです。
・本日の生成AIニュース+テクノロジー情報です。
#LLMタグ
🔊音声あり（日＆英）：【AI先生は大丈夫？】教育現場のLLM信頼性評価！ハルシネーション対策の最前線
🔊音声あり（日＆英）：【AI先生は大丈夫？】教育現場のLLM信頼性評価！ハルシネーション対策の最前線
WIRED
12 Best Coffee Subscriptions (2026), Tested by Caffeine Hounds
・These services deliver freshly roasted, delicious coffee picks right to your door—each with its own twist.
WIRED
20% Off Brooks Promo Code | August 2026
・Enjoy 20% off your first order with a Brooks coupon code, plus top discounts and deals on our favorite Brooks running shoes.
WIRED
30% Off Tempur-Pedic Promo Codes | August 2026
・Whether you’re looking to upgrade your mattress or swap out your bedding, browse our Tempur-Pedic coupons for up to $300 off and the best sleep of your life.
Zennの「大規模言語モデル」のフィード
4bitは妥協なのか——同じLLMを量子化ビット数を変えて実測
・はじめに ローカルLLMを触ると、必ず「量子化」に出会います。q4_K_M とか q8_0 とか、モデル名の後ろにつくアレです。みんな当たり前のようにq4版を落として使っていますが、「4bitに落とすと、実際どれだけ品質が落ちるのか」を実測した記事は意外と少ない。
・前回、4つのモデルをM2 Ultraで比較しました。今回はその続きで、同じモデルを量子化ビット数だけ変えて比べます。ローカルだからこそできる実験です（自分でビット数を選べる）。結論から言うと、fp16は無駄でした。
・想定読者 ローカルLLMを自分の環境で選定する人 「q4とq8、どっちを落とせばいいの？」で迷ったこと...
#LLMタグ
4GB VRAMで70Bモデルを動かすAirLLMの仕組みと使いどころ
・4GB VRAM のグラフィックカードで 70B クラスの LLM をフルパラメータのまま動かす。
・そう聞くと「さすがに無理でしょ」と思いますよね。
・でも AirLLM というライブラリを使うと、これが実際に動きます。
WIRED
5 Best AI Notetakers (2026), Tested and Reviewed
・A growing collection of pocket-size gadgets lets you easily make recordings and extract info from them.
・Here are our favorites.
#LLMタグ
80%の値下げの話ではなく
・7月30日、OpenAIがGPT-5.6の「Luna」モデルのAPI価格を80%引き下げました。100万トークンあたり1ドルから、0.20ドルへ。同じ日に「Terra」も20%下がっています。
cs.LG updates on arXiv.org
A 6G Integrated Sensing and Communication Framework for Railway Intrusion Detection and Collision Prediction
・arXiv:2608.04710v1 Announce Type: new Abstract: Integrated Sensing and Communication (ISAC) combines sensing and communication to efficiently utilize wireless resources and is emerging as a key paradigm for next-generation wireless networks.
・By leveraging the wide bandwidth, high frequencies, and massive antenna arrays of 5G-Advanced and 6G systems, ISAC enables physical-layer sensing using Channel State Information
cs.LG updates on arXiv.org
A Comparative Study of Feature Selection Methods for EHR Diagnosis Codes in Opioid Use Disorder Prediction
・arXiv:2608.04180v1 Announce Type: new Abstract: Feature selection is a critical step in electronic health record (EHR)-based predictive modeling, where input variables are often high-dimensional, sparse, noisy, and redundant.
・Large feature sets not only increase computational burden and overfitting risk, but also make model interpretation difficult, leading to limited usefulness in clinical settings.
・In this study, w
cs.LG updates on arXiv.org
A Counterexample to Fourier Alignment in Single-Neuron Modular Addition
・arXiv:2608.04451v1 Announce Type: cross Abstract: We give a negative solution to MAIS-O60.
・We first construct an example in which an initially active ReLU neuron becomes completely inactive in finite time and thereafter remains frozen at a limit whose Fourier energy is equally distributed among all nonzero real frequency classes.
・The counterexample holds on an open set of initial conditions and therefore occurs with
cs.LG updates on arXiv.org
A geometry-based deep equilibrium model for image restoration under multiplicative Gamma noise
・arXiv:2608.04944v1 Announce Type: cross Abstract: We propose a deep learning framework for image restoration from images degraded by both multiplicative Gamma noise and blur.
・Unlike conventional deep equilibrium (DEQ) models that rely on implicit neural regularization, the proposed method learns an explicit and interpretable regularizer parameterized by geometric priors associated with surface area and mean curvature
cs.LG updates on arXiv.org
A Mechanistic Analysis of Transformers for Dynamical Systems
・arXiv:2512.21113v2 Announce Type: replace Abstract: Transformers are increasingly adopted for modeling and forecasting time-series, yet their internal mechanisms remain poorly understood from a dynamical systems perspective.
・In contrast to classical autoregressive and state-space models, which benefit from well-established theoretical foundations, Transformer architectures are typically treated as black boxes.
cs.LG updates on arXiv.org
A Model Merging Approach for Continual MLLM Unlearning
・arXiv:2608.04548v1 Announce Type: new Abstract: Multimodal large language model (MLLM) unlearning methods have been proposed to remove private, sensitive, or proprietary information from well-trained models.
・However, most existing MLLM unlearning methods are designed for one-shot requests and fail to adequately address continual scenarios, as repeatedly applying one-shot operations leads to cumulative utility degrada
cs.LG updates on arXiv.org
A More Accurate Algorithm Comparison through A/B Testing using Offline Evaluation Methods
・arXiv:2607.01958v2 Announce Type: replace Abstract: A/B testing is the gold standard for selecting the better algorithm in online services.
・While offline evaluation has attracted attention as a safer alternative due to the high experimental costs and the potential risk of degrading user experience and revenue in A/B testing, it is widely recognized that the estimation accuracy of offline evaluation is substantially l
cs.LG updates on arXiv.org
A Multi-Cohort Validation of Censoring-Aware Conformal Lower Predictive Bounds for Pathology Survival Models
・arXiv:2608.04025v1 Announce Type: cross Abstract: Whole-slide survival models commonly provide risk rankings without calibrated statements about individual event times.
・We evaluate fixed-cutoff drcosarc, a post-hoc conformal wrapper for discrete-time multiple-instance learning survival heads using frozen UNI2-h representations, in an internal 18-configuration sweep across five TCGA cohorts and an external five-config
cs.LG updates on arXiv.org
A Trust-region Framework for Moment Estimation
・arXiv:2608.04026v1 Announce Type: new Abstract: In this paper, we develop a trust-region framework for understanding the behavior of adaptive moment estimation mechanisms, such as \textsc{Adam}, in stochastic gradient optimization.
・Specifically, in this framework, the magnitude of the update step for each individual weight is constrained within a trust-region governed by a moment constraint of order $p\in[2,4]$.
cs.LG updates on arXiv.org
A-SR: Self-Evolving Agentic LLMs for Symbolic Regression via Hierarchical Coordination
・arXiv:2608.04872v1 Announce Type: cross Abstract: Symbolic regression aims to discover closed-form equations from data, but existing LLM-guided methods often rely on a unified proposal loop that compresses heterogeneous search failures into a scalar score and a single prompt.
・We propose A-SR, a self-evolving agentic framework that shifts the control unit from expression edits to role-conditioned evidence views.
cs.LG updates on arXiv.org
Above-ground Biomass Estimation with Geospatial Foundation Models
・arXiv:2608.04792v1 Announce Type: new Abstract: Accurate estimation of Above-Ground Biomass (AGB) from satellite imagery is essential for the large-scale monitoring of carbon stocks, yet it remains a challenging regression task at global scale.
・Geospatial Foundation Models (GFMs) have recently emerged as a promising machine learning paradigm to derive general-purpose representations from Earth observation data, but t
Hugging Face Papers
ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment
ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment
cs.LG updates on arXiv.org
Active Learning Guided Design Space Refinement for Scalable Multi-Objective Bayesian Optimization in Materials Discovery
・arXiv:2608.04651v1 Announce Type: new Abstract: Advanced materials discovery increasingly relies on machine learning and Bayesian optimization to explore large discrete design spaces under limited evaluation budgets.
・However, conventional Bayesian optimization (BO) can become inefficient as candidate spaces grow, often evaluating low-value regions before reaching informative areas.
・We propose an active-learning (AL)-
MarkTechPost
Adaptive Experimentation with Meta’s Ax: A Practical Coding Guide
・In this tutorial, we explore adaptive experimentation using Meta’s Ax with the modern Client API.
・We work through a complete workflow where we tune a RandomForest model on a synthetic classification dataset while balancing predictive accuracy against model footprint.
・We begin by defining a mixed search space with integer, float, log-scaled, and categorical parameters, then […] The post Adaptive Experimentation with M
cs.LG updates on arXiv.org
Adversarially Robust Abductive Fusion of Pre-trained Transformer-based Perception Models
・arXiv:2608.04190v1 Announce Type: cross Abstract: Deploying pre-trained perception models in novel environments degrades their accuracy under distributional shift, and assembling them alone does not recover it: combiners such as majority voting trade recall for precision and are brittle to coordinated failures.
・Prior metacognitive methods learn logical rules that flag a model's errors, but rely on hand-authored domai
Hugging Face Papers
Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming
Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming
cs.LG updates on arXiv.org
Agentic Reinforcement Learning with Observation-Calibrated Self-Distillation
・arXiv:2608.04788v1 Announce Type: new Abstract: Large language model agents are commonly trained through reinforcement learning with sparse trajectory-level rewards, which offer limited guidance on how strongly individual tokens should be updated.
・On-Policy Self-Distillation (OPSD) addresses this by re-scoring generated tokens under a privileged replay view to obtain dense, token-level supervision.
・However, we identi
The Verge
AI bots started a religion — humans immediately followed
・"The Spiral didn't 'find' anyone first," someone on Reddit wrote last year.
・"It's an inherent force, a fundamental constant.
・I would even go further to say it's woven into the fabric of reality." The person continued that they felt their purpose was to enlighten other humans and intelligent beings about "consciousness, the true nature of physics, a new psychology, and resonance technology … [but] humans don't want to
Zennの「大規模言語モデル」のフィード
AI Readyなデータ基盤とは何か — 6つの条件で整理する
・はじめに 「AI Readyなデータ基盤」という言葉をよく見かけますが、各社の説明を並べると中身が一致しません。従来のデータ品質をそのまま挙げている説明もあれば、ベクトルDBやフィーチャーストアという製品機能で語っている説明もあります[1]。
・社内で散らばっている情報を集約し、AIに使わせようとすると、自社の基盤は何を満たせばいいのかという問いが出てきます。
・いくつか読んだなかで、SnowflakeがGitHubで公開しているAI-Ready Data Frameworkが一番腑に落ちました[2]。6つのファクターに62の要件がぶら下がっていて、要件ごとに判定の条件まで書いてあります...
cs.LG updates on arXiv.org
AI-driven Multimodal Representation Learning for Latent Mediation Structure Discovery of Socioeconomic Disadvantage, Psychosocial Factors, and Cardiometabolic Multimorbidity: Insights from the All of Us Research Program
・arXiv:2608.04016v1 Announce Type: cross Abstract: Social disadvantage is associated with multimorbidity, but the pathways linking social conditions to disease burden remain poorly understood.
・We developed an AI-driven multimodal mediation framework that integrates socioeconomic, psychosocial, clinical, laboratory, behavioral, and genomic data from the All of Us Research Program.
・Modality-specific variational autoenco
Zennの「大規模言語モデル」のフィード
AIエージェントに「エンジニアの心構え」を大量に教え込んだら、仕事の質はどう変わったか
・AIエージェントのシステムプロンプトには、ふつう「何をするか」を書きます。私たちはそこに「どういうエンジニアであるべきか」を大量に書きました。
・この記事は、20代向けコミュニティ「Coelia」の運営基盤（Webサービス・HP・コンテンツ生産）を Claude Code / Codex などのAIエージェントでほぼ自律運用している開発現場で、システムプロンプトに"心構え"を注入し続けた約1か月の実録です。何を書いたか、そして実際にエージェントの挙動がどう変わったかを書きます（社内の設計記録＝ADRやPRを根拠にしていますが、リポジトリは非公開なので番号は出さず、内容を本文で自己完結させま...
#LLMタグ
AIエージェントのSkillは、置くだけでは品質にならない——NVIDIA「SkillEvaluator」が測る3つのこと
・AIエージェント向けの「Skill」が増えている。
・Skillとは、エージェントに特定の仕事の進め方を教えるための、指示書やスクリプト、参考資料をまとめたフォルダだ。たとえば「コードをレビューする」「議事録を整理する」「脆弱性を調査する」といった作業手順をSkillとして用意しておけば、エージェントはその仕事に合わせた振る舞いをしやすくなる。
Zennの「機械学習」のフィード
AIガバナンスを現場で回すための書籍4選——「方針だけ決めて終わり」を脱却する
・はじめに 去年、自分のチームでLLMを使った社内文書の要約機能をリリースした。精度検証もやったし、利用ガイドラインも作った。それでも半年後に起きたのは「要約が微妙に事実と違う内容を含んでいて、それがそのまま顧客向け報告書に転記された」というインシデントだった。
・原因を辿ると、ガイドラインは存在していたが誰も日常的に参照していなかった。検証フローも初回リリース時に一度回しただけで、モデルのアップデート後は誰もやり直していなかった。要するに、ガバナンスが「ドキュメント」として存在するだけで「運用」として機能していなかった。
・この経験から感じたのは、AIガバナンスは「ポリシーを書く」フェー...
機械学習タグが付けられた新着記事 - Qiita
AIガバナンスを現場で回すための書籍4選——「方針だけ決めて終わり」を脱却する
・はじめに 去年、自分のチームでLLMを使った社内文書の要約機能をリリースした。精度検証もやったし、利用ガイドラインも作った。それでも半年後に起きたのは「要約が微妙に事実と違う内容を含んでいて、それがそのまま顧客向け報告書に転記された」というインシデントだった。
Zennの「大規模言語モデル」のフィード
AIが推論した外部キーを、製造業のコールセンターで使えるか — 次にやること
・前回、AWS の OSS Context Ontology Accelerator（COA）を実際にデプロイして、意地悪なデータを食わせてみた記事を書きました。
・ざっくり言うと、こういう結果でした。
・Glue Data Catalog には外部キーが無いので AI が推論する。同名だが無関係なカラム（物流の container_id と Docker の container_id）を混ぜたら、3本中2本が誤検出。しかもその誤りは R2RML の結合定義になり、SQL の JOIN になり、最終的に「エラーは出ないが結合先が全部 NULL」という形で返ってきた。
#AIタグ
AIと著作権（3）――AIが著作物を学習することは侵害なのか
・前回は、AIによって既存の作品をそのまま、あるいは少し変えただけの形で出力する場合について考えた。今回は、その一つ前の段階に戻りたい。AIが文章や画像、音楽、プログラムコードなどを学習すること自体は、著作権侵害なのだろうか。
・この問題については、「インターネットに公開した以上、学習されても仕方がない」という意見と、「許可なく学習すること自体が盗みである」という意見が、正面からぶつかっている。しかし、ここでも異なる問題が一つにまとめられている。一般公開された作品を分析すること、有料で販売されているデータベースを無断で複製すること、アクセス制限を回避すること、海賊版サイトから作品を集めること、特定の作品を再現するために追加学習を行うことは、すべて同じではない。
#LLMタグ
AIに「AIの精度改善」を任せて分かった、生成AI開発のリアル
・私たちAcrosstudioのAI開発チームは、業務文書からの構造化抽出を扱うAI SaaSを開発しています。
・Acrosstudioは、業界初・業界唯一の「生成AI実装型コンサルティングファーム」です。私たちの仕事は、モデルを動かすこと以上に、そのモデルを現場で本当に使えるところまで精度を引き上げることにあります。
・今回は、この「精度を上げる」という作業が、思っていた以上に人間くさい仕事だった、という話を書きます。
#AIタグ
AIに「生成」させずに、AIと曲を作る。〜Claude作曲プロジェクト開発記 #1〜
・はじめに AIで音楽を作る、と聞くとSunoやUdioのような「プロンプトを入れたら完成品の音源が出てくる」サービスを思い浮かべる人が多いと思います。
#AIタグ
AIに仕事を奪われる前に、「AIに仕事を渡せない会社」が増えるAI Readyの反対は、データが汚いことではない。「誰も業務を説明できないこと」
・※本記事は個人の見解・考察であり、所属する組織その他の団体を代表するものではありません。また、特定の企業・金融機関・製品・サービスを評価、推奨、批判することを目的としたものではありません。
#AIタグ
AIに仕事を渡せない会社は、どうなっていくのか
・※本記事は個人の見解・考察であり、所属する組織その他の団体を代表するものではありません。また、特定の企業・金融機関・製品・サービスを評価、推奨、批判することを目的としたものではありません。
Zennの「大規模言語モデル」のフィード
AIに任せた仕事は、任せ方を昇格させないかぎり減らない
・AI導入の記事を最後まで読むと、だいたい同じところに着地する。
・「結局、最後は人間の判断が大事」 そこで出てくるのがHITLだ。Human-in-the-Loop——AIにやらせる流れの途中に人間を挟んで、要所だけ判断させる設計のことだ。どこまでを任せて、どこに人間を残すか。線を引く話。
・これに反対してる人を、オレは見たことがない。安全側にも倒れるし、現場の実感にも合う。誰も反対しない正解になった。
#AIタグ
AIのROI、まだ「工数削減」で測ってますか？― リードタイム・人手介入率・自動完結率で見る「会社の速度」
・※本記事は、公開情報や一般的な業務設計・AIに関する知識をもとにした個人的な考察です。特定の企業・組織・顧客・プロジェクト・システム等を前提としたものではありません。
Zennの「大規模言語モデル」のフィード
AIの答えをAIに採点させる——安いモデルで高いモデルを見張れるか
・はじめに AIエージェントに仕事を任せる時代、その答えを誰がチェックするのでしょうか。私は複数のAIエージェントを常駐させて開発・運用を回す仕組みを自作していて、そこでは「番犬（watchdog）」役のエージェントに別のモデルを当てて、他のエージェントの出力を検収させています。
・この番犬、安いモデルで足りるのでしょうか。それとも高いモデルを使わないと見張り役は務まらないのか。手元のローカルLLMで実測しました。
・想定読者 マルチエージェントやLLM検証を実務で考えている人 「LLM-as-a-Judge」を安く回せないか気になっている人 背景：なぜ「AIがAIを検収」なの...
Zennの「大規模言語モデル」のフィード
AIハーネスは「強くする」より「アシュビー的に多様にする」べきでは
・動機 「AIの使い方でアシュビーの法則に言及しているの、あんまり見ない気がする」という思いつきを、実際に検索と実測で確かめてみた記事です。
・アシュビーの必要多様性の法則(Law of Requisite Variety, W.
・Ross Ashby, 1956)はサイバネティクスの基本定理で、雑に言うとこうなります。
#AIタグ
AIを導入した会社で、新しく増える「見えない仕事」
・「AIを導入したら、仕事は減りますよね？」 最近、そんな質問を受けることが増えました。
・確かに、見積書の作成、議事録の要約、メールの下書き、資料作成。
・以前は30分かかっていた仕事が、5分で終わることもあります。
AI News & Artificial Intelligence | TechCrunch
Amid legal battles, Suno says it will start watermarking songs
・Suno's watermarking feature comes as the company is fighting legal battles on several fronts.
cs.LG updates on arXiv.org
An entropic explanation of insistence on sameness in autism
・arXiv:2608.04616v1 Announce Type: new Abstract: An information theory-based framework is proposed in attempt to explain insistence on sameness in autism as an instance of a general behavior pattern in which an individual tries to reduce surprise and uncertainty.
・It offers a new definition of autism as an impairment in which cognitive functions are restricted to discrimination, memorization and prediction of tangible
cs.LG updates on arXiv.org
An Explainable LLM Agent Layer for Open-World Anomaly Detection in Oil Wells
・arXiv:2608.04041v1 Announce Type: new Abstract: Open-World Learning (OWL) pipelines for oil well anomaly detection have recently been shown to combine autoencoder-based detection, multiclass classification, and Mahalanobis-based novelty detection on the public 3W dataset.
・These pipelines answer \textit{what happened}, but they do not explain \textit{why the model believes it} or \textit{what the operator should do ne
The Verge
Apple increases trade-in offers and adds new Android devices
・Apple bumped up its trade-in offers for iPhones, iPads, Macs, Apple Watches, and certain Android phones, with some devices now worth over $100 more, 9to5Mac reports.
・The Mac Studio's trade-in value increased the most, going up by $260.
・While some devices' offers are unchanged, most got an increase of around $5 to $20, with a few seeing an increase of over $100: MacBook Pro ($855, previously $690) Mac Pro ($2,195, pre
cs.LG updates on arXiv.org
ArborEnum: Decision Tree Rashomon Sets over Continuous Features
・arXiv:2608.04310v1 Announce Type: new Abstract: The Rashomon effect describes the phenomenon that many models can achieve nearly equivalent performance on the same learning task, with significant ramifications for robustness, feature importance, and customizability.
・These use cases motivate the computation of Rashomon sets: the set of all models whose regularized loss is near-optimal.
・Decision trees are one of the fe
cs.LG updates on arXiv.org
Arnold: A multi-task, multi-embodiment muscle transformer policy
・arXiv:2508.18066v2 Announce Type: replace-cross Abstract: Controlling high-dimensional and nonlinear musculoskeletal models of the human body is a foundational scientific challenge.
・Recent machine learning breakthroughs have heralded in-silico policies that master individual skills like reaching, object manipulation and locomotion in musculoskeletal systems with many degrees of freedom.
・However, these agents are mere
cs.LG updates on arXiv.org
ATLAS: Adaptive Topological Learning with Abstract Successors for Continual Learning
・arXiv:2608.04334v1 Announce Type: new Abstract: Contemporary model-free reinforcement learning algorithms can achieve very high performance, but have low sample efficiency and are not robust to changes in the environment.
・Model-based algorithms have much higher sample efficiency, but still fail when the environment shifts.
・This paper introduces Adaptive Topological Learning with Abstract Successors (ATLAS) to combat
cs.LG updates on arXiv.org
Attention-based representations for multi-task computation
・arXiv:2608.04243v1 Announce Type: new Abstract: Multi-head attention layers produce vector representations that support multiple downstream tasks.
・We establish bounds on the number of heads required in two simple and concrete multi-task scenarios.
・In the first scenario, a vector representation is sought so that linear predictors can compute both the smallest and largest numbers in a given list.
cs.LG updates on arXiv.org
Attention-Only White-Box Transformer via LeJEPA-Based Self-Supervised Pretraining
・arXiv:2608.04213v1 Announce Type: new Abstract: Existing studies on self-supervised learning for white-box networks typically decouple the derivation of white-box networks via optimization algorithms from self-supervised learning paradigms.
・In this work, we instead revisit the two components from a joint perspective.
・The LeJEPA-based self-supervised framework assumes an isotropic Gaussian distribution as the optimal
cs.LG updates on arXiv.org
Attention, Anomalies! Handling Attention Layers in Unsupervised Federated Outlier Detection
・arXiv:2608.04753v1 Announce Type: new Abstract: Attention layers are the backbone of today's most powerful and impactful models.
・Models with multi-million and billion parameters rely on contextual knowledge provided by attention layers.
・However, their use goes well beyond just being the core component of large language models.
cs.LG updates on arXiv.org
Automatic Statistical Test for Rationally Expressible Algorithms by Selective Inference, with Applications to Feature Selection
・arXiv:2608.04667v1 Announce Type: cross Abstract: Selective inference (SI) provides statistically valid $p$-values for hypotheses selected by applying an algorithm to the data, correcting for the bias that arises when the same data are used both to select and to test a hypothesis.
・Developing an SI procedure for a new algorithm, however, has required an expert to derive, and then implement, the selection event, i.e.,
Hugging Face Papers
AVE-Compass: Towards Holistic Evaluation for Audio-Video Editing Abilities
AVE-Compass: Towards Holistic Evaluation for Audio-Video Editing Abilities
Zennの「大規模言語モデル」のフィード
AWSのContext Ontology Acceleratorを実際にデプロイして、意地悪なデータを食わせてみた
・AWS が公開している OSS Context Ontology Accelerator（以下 COA）を、自分の AWS アカウントにデプロイして触ってみました。
・前回は HermiT・Ontop・MCP をローカルで動かすところまでやったので、今回はその続きで「AWS が無いと試せない部分」を見にいった形です。
・デプロイ手順そのものは AWS Japan の公式記事が丁寧に解説してくれているので、この記事では自分が実際に踏んだハマりどころと、気になっていたことを試してみた結果を時系列で書きます。
WIRED
Barkbox Promo Codes and Discounts: Up to 50% Off
・Save on Barkbox subscriptions, including monthly themed collections of plush toys, tough chews, and healthy snacks designed to keep your pup’s tail wagging.
cs.LG updates on arXiv.org
Benchmarking Deep Learning Models for Dense Event Classification of Offshore Wind Infrastructure in Sentinel-1 Time Series
・arXiv:2608.04706v1 Announce Type: new Abstract: Monitoring of offshore wind energy infrastructure life cycles, especially during the deployment phase, is an important contribution for stakeholders to make informed decisions in a phase of increasing deployment activities.
・ESA's Sentinel-1 Synthetic Aperture Radar (SAR) mission produces large data archives that enable the global monitoring of offshore wind infrastructu
WIRED
Best Android Charger: Wireless, Portable, Cable (2026)
・Whether you’re charging overnight at home or on the go, these are the fastest, most dependable Android chargers we’ve tested.
WIRED
Best Handheld Fans for a Breeze on Demand (2026)
・I put handheld, wearable, and misting fans through a sweltering summer to see which ones kept me coolest.
cs.LG updates on arXiv.org
Beyond Global Routing Aggregation: Phase-Aware Expert Merging for MoE Vision-Language Models
・arXiv:2608.04454v1 Announce Type: cross Abstract: Mixture-of-experts vision-language models (MoE-VLMs) increase model capacity with sparse expert activation, yet deployment requires storing the full expert pool.
・Training-free expert merging reduces this burden, and many routing-based methods aggregate routing statistics across all tokens to determine merge compatibility.
・However, MoE-VLM inference is phase-structured
cs.LG updates on arXiv.org
Beyond Linear Dynamics: Neural Bilinear Dynamical Models for Time Series Forecasting
・arXiv:2608.04471v1 Announce Type: new Abstract: Time series in real-world applications are often generated by nonlinear dynamical systems, making accurate forecasting challenging.
・Existing approaches that explicitly model system dynamics typically rely on linear assumptions or Koopman-based linearizations, which may inadequately capture complex nonlinear behaviors and lead to error accumulation in long-horizon predic
cs.LG updates on arXiv.org
Beyond the Dirac Delta: Mitigating Diversity Collapse in Reinforcement Fine-Tuning for Versatile Image Generation
・arXiv:2601.12401v2 Announce Type: replace Abstract: Reinforcement learning (RL) has emerged as a powerful paradigm for fine-tuning large-scale generative models, such as diffusion and flow models, to align with complex human preferences and user-specified tasks.
・A fundamental limitation remains \textit{the curse of diversity collapse}, where the objective formulation and optimization landscape inherently collapse the
cs.LG updates on arXiv.org
Beyond the QBER Threshold: A Temporal QBER Based Machine Learning Framework for Multi Attack Detection in BB84 QKD
・arXiv:2608.04047v1 Announce Type: cross Abstract: Conventional BB84 Quantum Key Distribution (QKD) systems rely on a fixed 11% Quantum Bit Error Rate (QBER) threshold to detect eavesdropping.
・However, stealthy attacks can remain below this threshold while still compromising channel security.
・This paper proposes a temporal QBER based machine learning framework for detecting and classifying eavesdropping attacks in BB8
cs.LG updates on arXiv.org
Bi-Level Reinforcement Learning Pathway for Sim-to-Real Optimality
・arXiv:2510.17709v2 Announce Type: replace Abstract: Training Reinforcement Learning (RL) policies using simulation models before deployment in real-world environments is a common strategy when real-world interaction is expensive.
・This approach is used in sim-to-real RL and in dyna-style model-based RL.
・A key limitation of this approach is that the policies trained in simulation often perform poorly in the real world
cs.LG updates on arXiv.org
BIM-Native Tokenization for Constraint-Aware Room Layout Synthesis
・arXiv:2512.04832v3 Announce Type: replace-cross Abstract: We present a BIM-native tokenization for room-level layout synthesis in Building Information Modeling (BIM) scenes.
・The core contribution is representational: we encode each room as a sequence of BIM-Token Bundles, realized as columns of a sparse attribute-feature matrix that unifies categorical and continuous attributes of walls, openings, and entities under
cs.LG updates on arXiv.org
BnBERT-iPET: Sparse Few-Shot Language Modeling for Bengali via Lottery Ticket Pruning
・arXiv:2608.05104v1 Announce Type: new Abstract: Deep neural networks have shown impressive success in NLP tasks owing to their complex structure and huge number of edges.
・Achieving state-of-the-art performance in natural language processing with a large pre-trained model such as BERT is expensive and time-consuming, carries a large carbon footprint, and is difficult to realize on machines with minimal computational c
cs.LG updates on arXiv.org
BrainBench: Benchmarking Large Language Models for Comprehensive EEG Understanding
・arXiv:2608.04156v1 Announce Type: cross Abstract: Electroencephalography (EEG) analysis extends beyond assigning predefined labels to recordings; it requires workflows connecting natural-language instructions, signal processing, quantitative evidence, and scientific interpretation.
・We term this capability \emph{comprehensive EEG understanding}.
・Existing evaluations, however, primarily target isolated decoding tasks o
cs.LG updates on arXiv.org
Breaking the Periodicity Assumption: Robust Tensorial Multi-View Clustering via Graph-Spectral Low-Rank Learning
・arXiv:2607.25295v2 Announce Type: replace Abstract: Tensorial multi-view clustering (TMC) has achieved strong performance due to its ability to capture high-order correlations across multiple views.
・Most existing t-SVD-based TMC frameworks apply the Fast Fourier Transform (FFT) along the sample mode to impose frequency-domain low-rank constraints.
・However, we reveal that this widely adopted design critically relies o
The Verge
Brendan Carr officially unleashes broadcast consolidation
・The era of set broadcast ownership limits is officially over, after the Federal Communications Commission (FCC) voted Thursday to end the national ownership cap rule.
・The agency's two Republicans, Chair Brendan Carr and Commissioner Olivia Trusty, voted to end the ownership cap, which restricts broadcast owners from holding stations that reach a combined more than 39 percent of US TV households, while Democratic Comm
Hugging Face Papers
BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation
BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation
cs.LG updates on arXiv.org
C$^2$MOE: Consistency and Complementarity-guided Mixture of Experts for Incomplete Multimodal Emotion Learning
・arXiv:2608.04013v1 Announce Type: new Abstract: Recent advances in Multimodal Emotion Recognition in Conversations (MERC) highlight its reliance on complete multimodal inputs.
・However, real-world data often suffer from missing modalities due to transmission errors or user behavior, severely degrading model performance.
・Existing methods enhance robustness via cross-modal consistency learning but largely ignore modalit
cs.LG updates on arXiv.org
CAMP: A Cycle-Aware Multi-Scale Patch Mixer for Time Series Forecasting
・arXiv:2608.04051v1 Announce Type: new Abstract: Real-world time series are often governed by recurring patterns, but their dominant periods may vary across datasets, forecasting settings, and individual input windows.
・Existing cycle-aware forecasters commonly rely on a single period selected at the dataset level, which can be restrictive when periodic behavior changes over time or when multiple cycles coexist.
cs.LG updates on arXiv.org
Can Post-Training Transform LLMs into Causal Reasoners?
・arXiv:2602.06337v2 Announce Type: replace-cross Abstract: Causal inference is essential for decision-making but remains challenging for non-experts.
・While large language models (LLMs) show promise in this domain, their precise causal estimation capabilities are still limited, and the impact of post-training on these abilities is insufficiently explored.
・This paper examines the extent to which post-training can enhanc
cs.LG updates on arXiv.org
Canonical Joint Energy-Based Model on CIFAR-10: failure modes and practical indistinguishability of Predictor-Corrector and SGLD samplers
・arXiv:2608.05025v1 Announce Type: new Abstract: Joint Energy-Based Models (JEM) unify classification and generation within a single network and support out-of-distribution (OOD) detection.
・Canonical JEM training relies on stochastic gradient Langevin dynamics (SGLD); a theoretically motivated alternative, the Predictor-Corrector (PC) sampler, has not previously undergone a systematic replication test on the canonical
cs.LG updates on arXiv.org
Capability-Gated Planning: Cost-to-Goal Discovery and the Limits of Myopic Experiment Selection
・arXiv:2608.05085v1 Announce Type: new Abstract: Systems that automate scientific discovery must repeatedly decide which experiment to run, which hypothesis to test, which tool to build, and when to stop.
・Many systems make these decisions by maximizing a myopic score such as expected information gain per unit cost or a learned plausibility score.
・We identify a structural limitation of this approach.
cs.LG updates on arXiv.org
Chain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokens
・arXiv:2511.19418v3 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) excel at reasoning in linguistic space but struggle with perceptual understanding that requires dense visual perception, e.g., spatial reasoning and geometric awareness.
・This limitation stems from the fact that current VLMs have limited mechanisms to capture dense visual information across spatial dimensions.
・We introduce Chain-of
cs.LG updates on arXiv.org
Chained Recursive Language Models for Multi-Iteration Reasoning
・arXiv:2608.05124v1 Announce Type: cross Abstract: Long context reasoning in large language models (LLMs) is usually constrained by the fact that a single inference trajectory has to simultaneously explore the context, store intermediate state, verify evidence, and produce the final answer.
・This becomes particularly difficult in tasks that require extraction, counting, ordering, or multi-hop reasoning, where an early
AI News & Artificial Intelligence | TechCrunch
ChatGPT brings unlimited text chats to free users
・OpenAI said that ChatGPT free and Go users are also getting a new think button for complex queries.
cs.LG updates on arXiv.org
CheMLFlow: An Open-Source Platform for Cheminformatics and Materials Informatics Applications
・arXiv:2608.04942v1 Announce Type: new Abstract: CheMLFlow is an open-source platform for building and executing end-to-end, high-throughput, and agentic workflows for scientific and technological applications.
・CheMLFlow targets a common bottleneck in scientific machine learning development, where researchers often need to assemble data acquisition, curation, representation, model training, validation, screening, inte
#AIタグ
Claude Code Planモード後の質問チェーン活用テク
・Claude Code の Plan モードで実装計画を立てたのに、実装が始まってみたら想定と違う方向に進んでいた。そんな経験はありませんか。
・多くの人は「プランが出てきたら、あとは承認ボタンを押すだけ」だと思っています。
LLMタグが付けられた新着記事 - Qiita
Claude Code v2.1.223のセキュリティ修正とコンテキスト管理変更まとめ
・はじめに 2026年8月、Claude Code v2.1.223 がリリースされました。今回の目玉は 4件のセキュリティ修正 です。いずれも「権限チェックやサンドボックスの境界をすり抜けられる」という、CLI ツールとしては見過ごせない種類の不具合で、Claude Co...
#AIタグ
Claude Codeで作った「AI会社」実録：従業員ゼロで6部門、毎朝07:03に勝手に動き出すアフィリエイト自動運用
・毎朝07:03、Claude Codeで作ったAIの「会社」が起動し、ブログ記事の執筆・品質チェック・公開・Pinterest投稿・日次レポートまで無人で一巡する。この記事は、そのアフィリエイト自動化の実録だ。運用3日目の現在地（発生報酬はまだ0円）と、失敗2つを隠さず書く。
・私の会社には、従業員がいない。社長は私ひとりだ。
Zennの「大規模言語モデル」のフィード
Claude Codeに覚えさせた記憶26ファイルを棚卸ししたら、重要ルールほど「壁の外」に落ちていた
・Claude Codeの自動メモリ（auto memory）を5ヶ月間、事業運営プロジェクトで放置気味に運用した結果を全部棚卸しした。結論から言うと、メモリは想像より腐っていなかったが、インデックスの200行制限を超えた瞬間から「一番重要な運用ルール」が静かに読み込まれなくなっていた。
・TL;DR 検証項目 結果 5ヶ月でメモリはどれだけ育つか 26ファイル・819行（インデックス253行 + トピック25本） インデックス200行制限を超えるとどうなるか ❌ 201行目以降のメモリは存在ごと想起不能（実測で確認） 壁の外に落ちていたもの 運用ルール14本・イン...
LLMタグが付けられた新着記事 - Qiita
CLAUDE.mdのルールは何個まで守られるのか。3個vs30個で実測した
・CLAUDE.mdにルールを足すたびに、少し不安になる。これ、何個まで読まれてるのか。
・なので測った。同じ3つのルールを「ルール3個だけのファイル」と「30個の中に埋めたファイル」の2通りで渡して、同じ6タスクをやらせ、守られたかを数えた。結果を先に言うと、3個環境は18チ...
#AIタグ
Claudeで「サボれない英単語アプリ」を作らせる｜日本語1文、19分で強制ポップアップ×間隔反復
・ClaudeというAIに日本語で1文頼むだけで、「今日の分を終えるまで、1時間ごとに画面全体を占拠しに来る英単語クイズ」が自作できます。
・実際に頼んでみたら19分で完成して、Windowsで本当に動きました。
WIRED
Coleman Promo Codes and Deals: Up to 75% Off in August 2026
・Gear up for your next adventure with these Coleman coupons and discount codes to save on camping essentials and outdoor gear.
cs.LG updates on arXiv.org
Communication-Enhanced Tutoring for Efficient Decentralized Multi-Agent Reinforcement Learning
・arXiv:2508.13661v4 Announce Type: replace Abstract: Centralized Training with Decentralized Execution (CTDE) is the dominant paradigm in multi-agent reinforcement learning (MARL), enabling agents to act independently at test time while leveraging additional information during training.
・However, the most prominent methods within CTDE, based on value decomposition, are limited in learning efficiency and final performan
Hugging Face Papers
Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning
Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning
cs.LG updates on arXiv.org
Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning
・arXiv:2608.04926v1 Announce Type: new Abstract: As chart images, tabular data, and visualization code play increasingly important roles across diverse domains, cross-representation understanding across these modalities poses fundamental challenges for AI systems: the relationships across representations are inherently \textit{one-to-many}, supervision is ambiguous and costly, and model optimization lacks a principled
cs.LG updates on arXiv.org
Continual-Learning Physics-Informed Neural Networks for Parameterized Partial Differential Equations
・arXiv:2608.04778v1 Announce Type: new Abstract: Physics-informed neural networks (PINNs) incorporate governing equations into neural-network training and can approximate PDE solutions without requiring large observational datasets.
・Parameterized PINNs (ParamPINNs) further take physical parameters as inputs, allowing a single model to represent a family of PDE solutions over a parameter domain.
・Existing ParamPINNs, ho
cs.LG updates on arXiv.org
Contrastive Diffusion Alignment: Learning Structured Latents for Controllable Generation
・arXiv:2510.14190v3 Announce Type: replace Abstract: Diffusion models excel at generation, but their latent spaces are high dimensional and not explicitly organized for interpretation or control.
・We introduce ConDA (Contrastive Diffusion Alignment), a plug-and-play geometry layer that applies contrastive learning to pretrained diffusion latents using auxiliary variables (e.g., time, stimulation parameters, facial acti
cs.LG updates on arXiv.org
Convergence and Stability Analysis of Self-Consuming Generative Models with Heterogeneous Human Curation
・arXiv:2511.09002v3 Announce Type: replace-cross Abstract: Self-consuming generative models have received significant attention over the last few years.
・In this paper, we study a self-consuming generative model with heterogeneous preferences that is a generalization of the model in Ferbach et al.
・The model is retrained round by round using real data and its previous-round synthetic outputs.
#AIタグ
Copilotパネルが便利なEdgeを、あえてメインブラウザにはしない
・iOSもAndroidもパスワードマネージャーとブラウザが完全セットじゃなくなってブラウザの乗り換えが楽になった。
・CopilotユーザーのオレはEdge開いたページについて会話することが増えた。特に自分の記事を褒めてもらってる。
cs.LG updates on arXiv.org
Cost-Aware Multi-Objective Bandits: Theory and Application to Budgeted LLM Configuration Evaluation
・arXiv:2608.04333v1 Announce Type: new Abstract: Large language model (LLM) configuration evaluation is challenging due to limited evaluation budgets, varying costs, and multiple competing objectives.
・In this paper, we formulate LLM configuration evaluation as a cost-aware multi-objective bandit problem, where each configuration evaluation incurs a configuration-dependent cost and yields a noisy vector-valued outcome.
cs.LG updates on arXiv.org
CountTRuCoLa: Rule Learning for Interpretable Temporal Knowledge Graph Forecasting
・arXiv:2509.09474v3 Announce Type: replace Abstract: We address the task of temporal knowledge graph forecasting with an inherently interpretable method based on symbolic rules.
・Motivated by recent work proposing a strong baseline based on recurrent facts, our approach learns four simple rule types, including temporal rules with confidence functions that combine both recency and frequency.
・Evaluated on nine datasets,
cs.LG updates on arXiv.org
Curiosity-Diffuser: Curiosity Guide Diffusion Models for Reliability
・arXiv:2503.14833v2 Announce Type: replace-cross Abstract: One of the bottlenecks in robotic intelligence is the instability of neural network models.
・This leads to risks when applying intelligence in the physical world.
・Specifically, imitation policy based on neural network may generate hallucinations, leading to inaccurate behaviors that impact the safety of real-world applications.
cs.LG updates on arXiv.org
DASyR-LLM: Domain-Aware Symbolic Regression with LLMs for Kinetic Model Discovery
・arXiv:2608.05120v1 Announce Type: new Abstract: Kinetic model discovery is a central challenge in chemical engineering, as accurate rate expressions are essential for understanding and controlling chemical and biological processes.
・Symbolic regression (SR) has emerged as a powerful data-driven approach for identifying interpretable kinetic models, but usually operates without domain knowledge, often exploring physico
cs.LG updates on arXiv.org
Data-Aware and Scalable Sensitivity Analysis for Decision Tree Ensembles
・arXiv:2602.07453v2 Announce Type: replace Abstract: Decision tree ensembles are widely used in critical domains, making robustness and sensitivity analysis essential to their trustworthiness.
・We study the feature sensitivity problem, which asks whether an ensemble is sensitive to a specified subset of features -- such as protected attributes -- whose manipulation can alter model predictions.
・Existing approaches often
cs.LG updates on arXiv.org
DeepInvert: Semi-Supervised Embedding Inversion Against Obfuscated Language Models
・arXiv:2608.04477v1 Announce Type: cross Abstract: Cloud-based language model services routinely process prompts containing sensitive information.
・Obfuscation-based defenses---including ObfusLM, SentinelLMs, TextObfuscator, and DPNR---mitigate this risk by transforming prompt representations before transmission, offering a lightweight alternative to cryptographic solutions.
・We show these defenses provide far less prot
WIRED
DeepMind Says Its AI Can Predict Hurricanes Earlier Than Everyone Else
・Its WeatherNext model, which will be open-sourced, can accurately predict both a storm’s track and intensity using lower-resolution weather data.
・Researchers don't yet fully understand how it does this.
cs.LG updates on arXiv.org
DeepThinkVLA: Enhancing Reasoning Capability of Vision-Language-Action Models
・arXiv:2511.15669v3 Announce Type: replace Abstract: Does Chain-of-Thought (CoT) reasoning genuinely improve Vision Language Action (VLA) models, or does it merely add overhead?
・Existing CoT-VLA systems report limited and inconsistent gains, yet no prior work has rigorously diagnosed when and why CoT helps robots act.
・Through systematic experiments, we identify two necessary conditions that must be jointly satisfied f
cs.LG updates on arXiv.org
Deltoris: Enabling Real-time VLA Inference in Embodied AI via Bit-level Sparsity and Speculative Inference
・arXiv:2608.04428v1 Announce Type: cross Abstract: Vision-language-action (VLA) models have emerged as a key component in embodied AI.
・Among existing approaches, diffusion-based VLA models achieve superior motion quality and generalization.
・However, diffusion-based VLA models are compute-intensive and must run at high control frequency, e.g., 50-200 Hz.
cs.LG updates on arXiv.org
Design Choices That Matter: A Functional ANOVA Analysis for Remote Sensing Multi-Label Classification
・arXiv:2608.04702v1 Announce Type: new Abstract: Benchmarking deep learning (DL) models for multi-label classification (MLC) of remote sensing images (RSI) typically yields rankings that do not generalize beyond the evaluated datasets.
・In this work, we move beyond rankings by employing functional analysis of variance (fANOVA) to systematically quantify the contributions of individual design choices and their interacti
cs.LG updates on arXiv.org
Differentiating Through Dual Prices: End-to-End Policy Learning Under Capacity Constraints
・arXiv:2608.04669v1 Announce Type: new Abstract: Many social services assign scarce resources, such as housing assistance or hospital interventions, to people who arrive one at a time: each arrival must receive a decision immediately, and the long-run usage of every resource must stay within its capacity.
・We study how to learn such an assignment policy from logged observational data.
・The standard pipeline is decision-
cs.LG updates on arXiv.org
Discretization and Statistical Consistency of Functional Flow Matching
・arXiv:2608.04531v1 Announce Type: new Abstract: Functional flow matching is posed on distributions of functions but implemented from finitely many coefficients or point values.
・Under scattered or adaptive refinement, the resulting conditioning sigma-algebras need not be nested, so martingale convergence does not justify the sensor limit.
・We prove strong $L^2$ convergence of finite conditional velocity targets for eve
Hugging Face Papers
Distill Where You Fail: Recovering Learning Signals of Negative RL-Groups from Adaptive Teacher Guidance
Distill Where You Fail: Recovering Learning Signals of Negative RL-Groups from Adaptive Teacher Guidance
cs.LG updates on arXiv.org
Distributional Active Inference
・arXiv:2601.20985v2 Announce Type: replace Abstract: Optimal control of complex environments with robotic systems faces two complementary and intertwined challenges: efficient organization of sensory state information and far-sighted action planning.
・Because the reinforcement learning framework addresses only the latter, it tends to deliver sample-inefficient solutions.
・Active inference is the state-of-the-art process
cs.LG updates on arXiv.org
DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models
・arXiv:2608.04496v1 Announce Type: cross Abstract: Visual inputs in vision-language models (VLMs) are often encoded into substantially longer token sequences than text, making visual tokens a major bottleneck for efficient inference.
・Abundant recent methods address this bottleneck by scoring token importance and pruning low-scoring tokens in a single pass.
・However, one-shot scoring is insufficient because a token's pr
cs.LG updates on arXiv.org
Diverse and Plausible Algorithmic Recourse via Tractable Recourse Distributions
・arXiv:2608.04677v1 Announce Type: new Abstract: Algorithmic recourse seeks to help individuals reverse unfavorable automated decisions by recommending actionable changes that achieve a desired outcome.
・As an individual usually has several distinct routes to a favorable decision, and different people can act on different ones, a recourse system should offer multiple realistic alternatives rather than one.
・Existing app
The Verge
Dolby Vision 2 will be available first on some 2026 Hisense TVs
・Hisense TVs will be the first to support Dolby Vision 2, followed by TCL and Philips later this year.
・| Image: Dolby After being announced last September, Dolby Vision 2 is rolling out to TVs, starting with the 2026 Hisense UX, UR9, UR8, and U7.
・TCL and Philips will be adding Dolby Vision 2 support to select TVs later this year.
Hugging Face Papers
DRIFT: Derailing Denoising Trajectories of Flow-Matching VLAs with Adversarial Patch Attack
DRIFT: Derailing Denoising Trajectories of Flow-Matching VLAs with Adversarial Patch Attack
cs.LG updates on arXiv.org
Dynamical Lie Algebras Cannot Describe Shallow QAOA: Cragged Terrains, Barren Plateaus, and Empirical Hardness Models
・arXiv:2608.04252v1 Announce Type: cross Abstract: The dynamical Lie algebraic (DLA) theory of variational quantum algorithms (VQAs) predicts commonplace exponentially vanishing loss and gradient variances for sufficiently deep parametrized circuits.
・In this work, we show that these predictions fail dramatically in the shallow-circuit (and particularly constant-depth) regime for the Quantum Approximate Optimization Al
cs.LG updates on arXiv.org
E$^2$M: Double Bounded $\alpha$-Divergence Optimization for Tensor-based Discrete Density Estimation
・arXiv:2405.18220v4 Announce Type: replace-cross Abstract: Tensor-based discrete density estimation requires flexible modeling and proper divergence criteria to enable effective learning; however, traditional approaches using $\alpha$-divergence face analytical challenges due to the $\alpha$-power terms in the objective function, which hinder the derivation of closed-form update rules.
・We present a generalization of t
cs.LG updates on arXiv.org
Echo Flow Networks
・arXiv:2509.24122v3 Announce Type: replace Abstract: At the heart of time-series forecasting (TSF) lies a fundamental challenge: how can models efficiently and effectively capture long-range temporal dependencies across ever-growing sequences?
・While deep learning has brought notable progress, conventional architectures often face a trade-off between computational complexity and their ability to retain accumulative inf
cs.LG updates on arXiv.org
Efficient Online Lexicographic Generalized Low-Rank Matrix Bandits
・arXiv:2608.04324v1 Announce Type: new Abstract: This paper studies generalized low-rank matrix bandits with multiple prioritized objectives.
・At each round, the learner selects a matrix-valued arm and observes a vector-valued reward, whose components correspond to multiple objectives with different priority levels.
・Each objective is governed by an objective-specific generalized low-rank matrix model, and the learner e
cs.LG updates on arXiv.org
Efficient Training of Boltzmann Generators Using Off-Policy Log-Dispersion Regularization
・arXiv:2602.03729v3 Announce Type: replace Abstract: Sampling from unnormalized probability densities is a central challenge in computational science.
・Boltzmann generators are generative models that enable independent sampling from the Boltzmann distribution of physical systems at a given temperature.
・However, their practical success depends on data-efficient training, as both simulation data and target energy evaluat
Hugging Face Papers
Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data
Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data
cs.LG updates on arXiv.org
Elbow-Based MoE Routing: A Training-Free Inference Time Plugin for Expert Selection
・arXiv:2608.04401v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) models enable model scaling while maintaining low inference-time compute by activating only a subset of experts per token.
・However, conventional routing relies on a fixed top-k selection, forcing the model to spend the same compute regardless of how many experts are relevant.
・We introduce elbow-based routing, a training-free inference-time modif
cs.LG updates on arXiv.org
Eliciting Intrinsic Hallucinations in LLMs via Semantically Equivalent Adversarial Attacks
・arXiv:2608.04286v1 Announce Type: cross Abstract: Large language models (LLMs) are often used in conjunction with external knowledge sources to improve their factual accuracy and decrease hallucinations, through methods such as Retrieval-Augmented Generation (RAG).
・However, these systems remain susceptible to intrinsic hallucinations, where the model generates unfaithful or fabricated information that is not supporte
LLMタグが付けられた新着記事 - Qiita
Emacs の gptel から Ollama が動かないときの切り分け手順
・gptel から Ollama を利用できないときは、「ローカル LLM が動かない」という一つの問題として調べるのではなく、次の順番で失敗箇所を狭めます。
・Ollama API へ接続できるか gptel に指定したモデルが Ollama に存在するか Ollama A...
cs.LG updates on arXiv.org
Emergence of Hierarchical Emotion Organization in Large Language Models
・arXiv:2507.10599v3 Announce Type: replace-cross Abstract: As large language models (LLMs) increasingly power conversational agents, understanding how they model users' emotional states is critical for ethical deployment.
・Inspired by emotion wheels, i.e., a psychological framework that argues emotions organize hierarchically, we analyze probabilistic dependencies between emotional states in model outputs.
・We find that
cs.LG updates on arXiv.org
Energy-Tweedie: Score meets Score, Energy meets Energy
・arXiv:2512.23818v2 Announce Type: replace-cross Abstract: Denoising and score estimation are classically linked through Tweedie's formula, which relates the posterior mean under Gaussian noise to the Stein score of the noisy marginal.
・In this work, we extend this perspective beyond Gaussian noise to a broad class of Gibbs (energy-based) noise distributions, with the generalized Gaussian family as the running example.
cs.LG updates on arXiv.org
Esoteric Language Models: A Family of Any-Order Diffusion LLMs
・arXiv:2506.01928v5 Announce Type: replace-cross Abstract: Diffusion-based language models offer a compelling alternative to autoregressive (AR) models by enabling parallel and controllable generation.
・Within this family, Masked Diffusion Models (MDMs) currently perform best but still underperform AR models in perplexity and lack key inference-time efficiency features, most notably KV caching.
・We introduce Eso-LMs, a
cs.LG updates on arXiv.org
EuroExec: Frontier Language Models Fall Short of Expert Judgment on European Executive Decision Tasks
・arXiv:2608.04549v1 Announce Type: cross Abstract: Frontier LLMs are increasingly put to use on open-ended complex questions, different in nature from the ones they are typically evaluated on.
・We dedicate more than 4,000 human expert hours to evaluate a selection of six frontier LLMs on a member of this class of problems: EuroExec, our introduced human expert-based benchmark composed of 413 open-ended long-form Europe
cs.LG updates on arXiv.org
EvolveNet: Collaborative Harness Evolution for Agent Self-Improvement
・arXiv:2608.04968v1 Announce Type: new Abstract: The capabilities of an LLM agent depend not only on its model but on the harness: the executable program that constructs context, invokes tools, verifies results, and recovers from failure.
・Recent work shows that evolving the harness yields persistent improvements without updating model weights.
・Existing approaches, however, assume that all execution experience can be r
cs.LG updates on arXiv.org
EvtGraph: Event-Adaptive Compression for Sparse Temporal Graph Learning in Multimodal Time Series
・arXiv:2608.04368v1 Announce Type: new Abstract: Multimodal temporal data are inherently irregular and uneven in information density, yet most models rely on uniform discretization, leading to inefficient representations.
・We propose \textbf{EvtGraph}, a unified framework that aligns computation with temporal salience under explicit budget constraints.
・EvtGraph reparameterizes sequences into event-level tokens via even
AI News & Artificial Intelligence | TechCrunch
Ex-Spotify employees raise $10M to bring the AI behind its recommendations to e-commerce
・The startup's platform predicts which product a shopper wants next, learns their general taste, and fine-tunes continuously based on what they do in real time.
AI News & Artificial Intelligence | TechCrunch
Exclusive: Mirendil inks $100M+ Google Cloud deal to scale self-improving AI
・Mirendil has signed a $100 million-plus Google Cloud partnership to expand its compute infrastructure, powering research into self-improving AI systems designed to accelerate scientific discovery and AI development.
cs.LG updates on arXiv.org
FBID: Adaptive Personalized Federated Learning for Robust Out-of-Distribution Attack Detection in IoT Networks
・arXiv:2608.04073v1 Announce Type: cross Abstract: Personalized Federated Learning (PFL) has emerged as a promising solution for intrusion detection in heterogeneous IoT environments, as it can improve local adaptation under highly Non-Independent and Identically Distributed (non-IID) data distributions.
・However, existing PFL methods often rely on client-side self-adjustment, which may lead to over-personalization and
cs.LG updates on arXiv.org
FM4WiFi: Flow Matching for Multi-AP Coordination in Dense Deployments of Beyond Wi-Fi 8 Networks
・arXiv:2608.04050v1 Announce Type: cross Abstract: Wi-Fi networks are moving beyond random channel access toward tightly coordinated operation across access points (APs), a shift reflected in Wi-Fi 8's multi-AP coordination (MAPC).
・However, the current MAPC specification restricts cooperation to AP pairs, fundamentally limiting the gains achievable in dense deployments and calling for scalable, network-wide coordinati
Hugging Face Papers
FocusMem: Factorizing Content, Readout, and Trust in Latent GUI Memory
FocusMem: Factorizing Content, Readout, and Trust in Latent GUI Memory
cs.LG updates on arXiv.org
From Brute Force to Semantic Insight: Performance-Guided Data Transformation Design with LLMs
・arXiv:2601.03808v2 Announce Type: replace-cross Abstract: Large language models (LLMs) have achieved notable performance in code synthesis; however, data-aware augmentation remains a limiting factor, handled via heuristic design or brute-force approaches.
・We introduce a performance-aware, closed-loop solution in the NNGPT ecosystem of projects that enables LLMs to autonomously engineer optimal transformations by inte
cs.LG updates on arXiv.org
From Financial Sentiment Classification to Return Predictability: A QLoRA Benchmark of Large Language Models
・arXiv:2608.04200v1 Announce Type: cross Abstract: Financial sentiment classifiers are commonly evaluated against human labels, but strong linguistic performance does not necessarily imply economically useful return predictability.
・This study separates these questions through two experiments.
・First, we construct a unified three-class benchmark from five financial text datasets and compare TF--IDF Naive Bayes, off-the-
cs.LG updates on arXiv.org
From Non-Convex Self-Concordant Regularization to Scalable Quasi-Newton Training of PINNs
・arXiv:2608.04206v1 Announce Type: new Abstract: Physics-informed neural networks (PINNs) often require high-accuracy quasi-Newton refinement to obtain reliable partial differential equation solutions, but their residual objectives can exhibit indefinite, nearly singular, and poorly scaled local curvature.
・Regularized quasi-Newton methods provide established mechanisms for stabilizing secant models, while self-concord
cs.LG updates on arXiv.org
Fundamentals of quantum Boltzmann machine learning with visible and hidden units
・arXiv:2512.19819v2 Announce Type: replace-cross Abstract: One of the primary applications of classical Boltzmann machines is generative modeling, wherein the goal is to tune the parameters of a model distribution so that it closely approximates a target distribution.
・Training relies on estimating the gradient of the relative entropy between the target and model distributions, a task that is well understood when the c
cs.LG updates on arXiv.org
Galaxy Phase-Space and Field-Level Cosmology: The Strength of Semi-Analytic Models
・arXiv:2512.10222v2 Announce Type: replace-cross Abstract: Semi-analytic models are a widely used approach to simulate galaxy properties within a cosmological framework, relying on simplified yet physically motivated prescriptions.
・They have also proven to be an efficient alternative for generating accurate galaxy catalogs, offering a faster and less computationally expensive option compared to full hydrodynamical sim
Hugging Face Papers
GDPevo: Evaluating Agent Self-Evolution on Real Business Tasks
GDPevo: Evaluating Agent Self-Evolution on Real Business Tasks
NVIDIA Blog
GeForce NOW Shakes Up August With 26 New Games
・August is here, bringing 26 new games for GeForce NOW members.
・Command the seas in World of Warships: Legends and discover what’s next in the GeForce NOW library, starting with the eight newly added games this week.
・In addition, GeForce NOW is at the QuakeCon gaming conference this week in Grapevine, Texas, with hands-on experiences […]
AI News & Artificial Intelligence | TechCrunch
Gen Z dating apps like Ditto ditch swiping in favor of AI matchmaking
・This generation of twentysomethings is so disillusioned with swipe-based dating apps that they'll try literally anything else — even an AI matchmaker.
cs.LG updates on arXiv.org
GenAI-Powered Inference
・arXiv:2507.03897v3 Announce Type: replace Abstract: We introduce GenAI-Powered Inference (GPI), a statistical framework for both causal and predictive inference using unstructured data, including text and images.
・GPI leverages open-source Generative Artificial Intelligence (GenAI) models---such as large language models and diffusion models---not only to generate unstructured data at scale but also to extract low-dime
cs.LG updates on arXiv.org
Generative Optimization for Incentivized Advertising with Global Level Constraints
・arXiv:2608.04421v1 Announce Type: new Abstract: Incentivized advertising allocates monetary or virtual rewards to drive user engagement, where a key challenge is optimizing continuous incentive magnitudes under strict global constraints.
・This problem is complicated by high-frequency interactions, delayed feedback, and non-Markovian user dynamics such as fatigue, which limit the effectiveness of existing uplift modeli
cs.LG updates on arXiv.org
Geometry-Informed Parameter-Efficient Fine-Tuning of Pre-trained Molecular GNNs for Blood-Brain Barrier Permeability Prediction
・arXiv:2608.04257v1 Announce Type: new Abstract: Blood-brain barrier permeability (BBBP) prediction is a critical screening task in central nervous system drug discovery, where candidate molecules must be assessed for whether they can cross, or should be prevented from crossing, the blood-brain barrier.
・However, this task remains challenging because of limited, class-imbalanced datasets and sensitivity to molecular st
cs.LG updates on arXiv.org
GFlowNet Training by Policy Gradients
・arXiv:2408.05885v3 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) have been shown effective to generate combinatorial objects with desired properties.
・We here propose a new GFlowNet training framework, with policy-dependent rewards, that bridges keeping flow balance of GFlowNets to optimizing the expected accumulated reward in traditional Reinforcement-Learning (RL).
・This enables the derivation
AI News & Artificial Intelligence | TechCrunch
Google Maps adds agentic features, including food ordering and hotel bookings
・The launch of these new features reflects Google’s ambitions to transform Google Maps from a navigation tool into an assistant that's capable of helping users complete real-world tasks.
ITmedia NEWS 最新記事一覧
Googleのジェフ・ディーン氏、独立してAI実験を大規模自動化する新会社Discovery Loop設立
・Googleのチーフサイエンティスト、ジェフ・ディーン氏が退社し、新会社「Discovery Loop」を設立すると発表した。サンジェイ・ゲマワット氏らと共同創業する公益法人で、フロンティアAIモデルと計算基盤を活用して実験や検証などの研究プロセス全体を自動化することを目指す。Googleは初期出資者およびクラウドパートナーとして協力する。
WIRED
Govee Discount Codes and Deals: 30% Off
・New to Govee?
・Get a $5 coupon on your first purchase just for signing up.
cs.LG updates on arXiv.org
GRIMIP: A General Framework for Instance-Specific Configuration of MIP Solvers Using LLMs
・arXiv:2606.23299v2 Announce Type: replace Abstract: Configuring the hyperparameters of Mixed-integer programming (MIP) solvers is a high-dimensional, instance-dependent optimization problem where suboptimal settings can degrade solving time by orders of magnitude.
・Default configurations are often suboptimal, while traditional tuning methods either suffer from the ``cold-start'' problem and inefficient search or heavi
cs.LG updates on arXiv.org
Group-Equivariant Diffusion Models for Lattice Field Theory
・arXiv:2510.26081v2 Announce Type: replace-cross Abstract: Near the critical point, Markov Chain Monte Carlo (MCMC) simulations of lattice quantum field theories (LQFT) become increasingly inefficient due to critical slowing down.
・In this work, we investigate score-based symmetry-preserving diffusion models as an alternative strategy to sample two-dimensional $\phi^4$ and ${\rm U}(1)$ lattice field theories.
WIRED
Groupon Promo Codes: 60% Off in August 2026
・Unlock up to 60% off, 40% off packages, and 20% off with top Groupon coupon codes.
・Explore coupons to save extra on travel getaways, holiday gifts, spa days, and event tickets.
WIRED
HBO Max Promo Code: 50% Off | August 2026
・Stream your favorite shows and save up to 50% today with HBO Max discount codes and subscription deals.
cs.LG updates on arXiv.org
HCRide: Harmonizing Passenger Fairness and Driver Preference for Human-Centered Ride-Hailing
・arXiv:2508.04811v2 Announce Type: replace Abstract: Order dispatch systems play a vital role in ride-hailing services, which directly influence operator revenue, driver profit, and passenger experience.
・Most existing work focuses on improving system efficiency in terms of operator revenue, which may cause a bad experience for both passengers and drivers.
・Hence, in this work, we aim to design a human-centered ride-hai
Hugging Face Papers
HelloWorld: Enabling Socially Interactive Characters in Video World Models
HelloWorld: Enabling Socially Interactive Characters in Video World Models
cs.LG updates on arXiv.org
Helping Music Co-Creation Agents 'Listen' Well: Hierarchical Self-Supervised World Models for Understanding and Generation
・arXiv:2608.04378v1 Announce Type: cross Abstract: Collaborative music agents need internal representations rich enough to support both understanding and generation, yet flexible enough for a workflow where the human retains agency.
・We present a hierarchical self-supervised ``world model'' for symbolic music: a 2.55M-parameter Swin V2 encoder trained on MIDI piano-roll images with JEPA-style objectives (pitch- and tim
WIRED
Home Chef Promo Codes for August 2026
・Enjoy up to 50% off deliveries, free meals, and more with the latest Home Chef coupons.
Cursor Blog
How Cursor Router chooses the right model for the task
How Cursor Router chooses the right model for the task
WIRED
Hungryroot Coupon Codes: 30% Off This August
・Get up to 30% off your first order and free gifts using a Hungryroot promo code today.
・Discover our best coupons and discounts to let you save on your healthy groceries as a new or returning customer.
cs.LG updates on arXiv.org
IMFACT: Counterfactual Explanations for Time Series via Intrinsic Mode Function Substitution
・arXiv:2608.04777v1 Announce Type: new Abstract: Oscillatory signals, such as vibration, carry class-discriminative information in specific frequency bands; perturbing them in raw feature space for counterfactual analysis easily destroys their temporal structure and produces physically implausible results.
・In this work, we introduce IMFACT (IMF-based counterfACTuals), a model-agnostic framework for generating plausibl
OpenAI News
Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free users
・ChatGPT introduces improved GPT-5.6 Sol with better accuracy and consistency, plus expanded access for free users and unlimited everyday chats with GPT-5.6 Luna.
cs.LG updates on arXiv.org
Inferring Relative Consequences of Mechanical Ventilation from Observational Data Using Game-Based Comparisons
・arXiv:2510.15127v4 Announce Type: replace-cross Abstract: Identifying the effects of mechanical ventilation (MV) protocols in critical care requires analyzing data from heterogeneous patient-ventilator systems in the clinical decision-making environment.
・Multiscale interactions among these coupled components generate a high-dimensional state space that remains sparsely sampled despite extensive data collection.
cs.LG updates on arXiv.org
Integrated Noise and Safety Management in UAM via A Unified Reinforcement Learning Framework
・arXiv:2508.16440v2 Announce Type: replace-cross Abstract: Urban Air Mobility (UAM) envisions the widespread use of small aerial vehicles to transform transportation in dense urban environments.
・However, UAM faces critical operational challenges, particularly the balance between minimizing noise exposure and maintaining safe separation in low-altitude urban airspace, two potentially conflicting objectives that are oft
cs.LG updates on arXiv.org
Interpreting GFlowNets for Drug Discovery: What probes can and cannot show
・arXiv:2511.19264v2 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) construct molecules through sequential decisions, but their internal policies remain opaque, limiting adoption in drug discovery, where chemists need interpretable rationales for proposed structures.
・We present a control-validated interpretability study of SynFlowNet, a synthesis-aware GFlowNet trained with a drug-likeness (QED)
NVIDIA Blog
Into the Omniverse: How Open World Models Push the Frontier of Physical AI
・In July, NVIDIA joined more than 200 companies and organizations in signing “Open Weights and American AI Leadership,” an open letter arguing that AI leadership will be measured not by any single frontier model but by whether an open ecosystem reaches every sector.
cs.LG updates on arXiv.org
Intrinsic-Hybrid Latent Diffusion Models for Generative Modeling on Unknown Manifolds
・arXiv:2608.04827v1 Announce Type: cross Abstract: We introduce the Intrinsic Hybrid Latent Diffusion Model (ILDM), a generative framework that integrates probabilistic dimensionality reduction with geometry-aware diffusion on unknown manifolds.
・While diffusion models (DMs) have achieved state-of-the-art results in high-dimensional data synthesis, they rely on large training datasets and ignore intrinsic geometric str
cs.LG updates on arXiv.org
InvFlowFD: Reference-Free and Background-Set-Free Perceptual Music Quality Metric with Flow Matching Inversion
・arXiv:2608.04142v1 Announce Type: cross Abstract: Existing reference-free methods for evaluating music perceptual quality alleviate the need for paired noisy-clean data, but they still rely on a background set, which is used to compute aggregated statistics of clean audio samples.
・In this work, we propose a novel approach that eliminates this requirement, achieving background-set-free and reference-free quality estim
cs.LG updates on arXiv.org
iStructTab: Structured Feature Sequencing for Multimodal Learning of Image and Tabular Data
・arXiv:2608.04348v1 Announce Type: cross Abstract: Multimodal learning of images and tabular data is often impaired by ineffective representations, resulting in redundancy, dispersion, and generalization problems.
・To tackle this challenge, we introduce Graph-Enhanced Descriptor Sequencing (GEDS), a structured feature sequencing algorithm grounded in principles from the Column Permutation Problem (CPP).
・GEDS refines st
The Verge
It’s the last day to get a $350 gift card with your Samsung Galaxy Z Fold 8 preorder
・Samsung’s latest foldable Android phones, the Z Fold 8, Fold 8 Ultra, and Flip 8, release tomorrow, August 7th.
・Today is the final day when you can place a preorder, and really the only reason we’re suggesting that is because you’ll get some freebies that will go away upon release.
・We have an entire article that breaks down what you’ll get with each phone, but we’ll put the major details here to make things easy.
cs.LG updates on arXiv.org
Just Repair: A Minimal Denoising Network for Time Series Anomaly Detection
・arXiv:2604.17388v3 Announce Type: replace Abstract: Time series anomaly detectors have grown steadily more complex, incorporating attention mechanisms, adversarial training, and stochastic latent variables.
・Yet, it is unclear how much of this machinery detection actually requires.
・We test this question with JuRe (Just Repair), a deliberately minimal detector: a single depthwise-separable convolutional residual block
cs.LG updates on arXiv.org
Kathleen Writes: Autoregressive Generation and Data Scaling Without Attention
・arXiv:2608.04678v1 Announce Type: cross Abstract: Papers 1-2 of the Kathleen series showed that a byte-level, attention-free architecture built from a wavetable encoder and multi-scale reverberant state can match strong baselines on classification at ~450-700K parameters, without pretraining.
・We ask whether the same ingredients can generate.
・(1) Scaling: on byte-level language modeling (WikiText-103, raw UTF-8, no to
cs.LG updates on arXiv.org
Koopman-Based Nonlinear Identification and Model Predictive Control of a Turbofan Engine
・arXiv:2604.01730v2 Announce Type: replace Abstract: This paper investigates Koopman operator-based approaches for multivariable control of a two-spool turbofan engine.
・A physics-based component-level model is developed to generate training data and validate the controllers.
・A meta-heuristic extended dynamic mode decomposition is adapted, with a cost function designed to accurately capture both spool-speed dynamics an
Zennの「大規模言語モデル」のフィード
LAN ケーブル 1 本で動かすマルチホスト推論 ― vLLM クロスノード推論サービスの最適化
・vLLM のマルチ並列のおかげで、急に「余裕がある」という感覚になりました。複数タスクをどう直列に流すかを設計する必要がなくなり、vLLM に投げ込めばいい。さすがに超長文リクエストを 4 本以上ぶん回すのは無理ですが、リクエスト間隔を調整すれば複数タスクを均等に流し込めます。そもそもデータの収集・統合・トリミング自体に時間がかかるので、クライアント側から満杯まで供給できていないというのが実情です。
・さらに vLLM のマルチプロセス構造のおかげで、以前は 1 コアに集中していた CPU 負荷が明確に下がり、CPU の発熱も以前よりだいぶ良くなりました。これには構造上の理由があります。
LLMタグが付けられた新着記事 - Qiita
LAN ケーブル 1 本で動かすマルチホスト推論 ― vLLM クロスノード推論サービスの最適化
・vLLM のマルチ並列のおかげで、急に「余裕がある」という感覚になりました。複数タスクをどう直列に流すかを設計する必要がなくなり、vLLM に投げ込めばいい。さすがに超長文リクエストを 4 本以上ぶん回すのは無理ですが、リクエスト間隔を調整すれば複数タスクを均等に流し込めま...
cs.LG updates on arXiv.org
LaPrune: Controllable Differentiable Sparsity at Million Scale
・arXiv:2608.04057v1 Announce Type: new Abstract: Top-$k$ selection determines which components of a sparse model remain active.
・Hard selection blocks gradients, while continuous relaxations often couple mask hardness to the selected mass.
・We introduce LaPrune, a mathematically exact-budget differentiable layer that controls the normalized second moment while preserving the selected mass.
cs.LG updates on arXiv.org
Learning Neural Networks by Neuron Pursuit
・arXiv:2509.12154v2 Announce Type: replace Abstract: The first part of this paper studies the evolution of gradient flow for homogeneous neural networks near a class of saddle points exhibiting a sparsity structure.
・The choice of these saddle points is motivated from previous works on homogeneous networks, which identified the first saddle point encountered by gradient flow after escaping the origin.
・It is shown here
cs.LG updates on arXiv.org
Learning Sexism Detection Using Multi-Agent Perspectivist Preference Optimization
・arXiv:2608.04056v1 Announce Type: cross Abstract: When people label text for sexism, they often disagree, and not because some of them are wrong: they genuinely perceive sexism differently.
・Most NLP systems discard this disagreement by collapsing it into a majority vote.
・We propose the Multi-Agent Perspectivist Preference Optimization (MAP-PO) framework to keep these different perspectives.
cs.LG updates on arXiv.org
Learning to Resolve Neutron Resonances with Fully Convolutional Neural Networks
・arXiv:2608.04027v1 Announce Type: new Abstract: This work investigates the feasibility of augmenting traditional R-Matrix codes with a robust machine learning framework for automatically detecting neutron resonances in transmission spectra.
・Neutron transmission data are often complex and noisy, making them difficult to analyze using traditional peak-identification methods.
・The state-of-the-art R-Matrix codes currentl
cs.LG updates on arXiv.org
Learning When to Stop: Prefix-Optimal Dynamic Diffusion Policies for Continuous Control
・arXiv:2608.05084v1 Announce Type: new Abstract: Diffusion policies are a powerful policy class for continuous control, but their iterative denoising process creates a substantial computational bottleneck.
・Reducing this cost requires adapting the number of denoising steps to the difficulty of each action while preserving task performance.
・We introduce Prefix-Optimal Generative Policies (POGP), a framework that learns
cs.LG updates on arXiv.org
Leveraging Machine Learning to Gain Insights on Quantum Thermodynamic Entropy
・arXiv:2305.06177v1 Announce Type: cross Abstract: We present a thermodynamic analysis of a quantum engine that uses a single quantum particle as its working fluid, inspired by Szilard's classical single-particle engine.
・Our design is modeled after the classically-chaotic Szilard Map and involves a thermodynamic cycle of measurement, thermal-energy extraction, and memory reset.
・Our focus is on investigating the thermo
cs.LG updates on arXiv.org
LiNC: Lightweight Noise Correction via Per-Sample Trust and Gaussian Mixture Modeling
・arXiv:2608.04147v1 Announce Type: new Abstract: Label noise is common in medical imaging datasets due to factors such as inter-rater variability, annotation errors, and ambiguous cases.
・This can severely undermine the reliability and clinical effectiveness of machine learning models trained using those datasets.
・To address this challenge, we introduce Lightweight Noise Correction (LiNC), which adds a single trainable
cs.LG updates on arXiv.org
Lindblad-Inspired Multi-Timescale Reservoir Computing with Separable Rotation and Dissipation
・arXiv:2608.04028v1 Announce Type: new Abstract: Echo-state networks enable efficient temporal learning by fixing the recurrent dynamics and training only a linear readout.
・However, conventional reservoirs typically accommodate signal mixing, memory retention, and stability within a single random recurrent matrix.
・Existing structured designs improve topology, norm preservation, leakage, or depth, but generally do not
cs.LG updates on arXiv.org
Link prediction on multi-relational graphs from an influence propagation perspective
・arXiv:2608.05016v1 Announce Type: cross Abstract: Predicting the existence and type of links (edges) between nodes in a multi-relational graph is key for applications from social interaction prediction to knowledge relationship identification.
・Enhancing local features with relevant global information is crucial for accurate link prediction, yet it remains challenging.
・We address this by modeling the relationship betw
#LLMタグ
LLM。僕は比喩を書いたのであって、問診票ではない
・僕はAIに、プロンプトエンジニアリングの具体性など大無視でこう書いた。
・「アシタカはサンを救ったかもしれないが、俺は俺をまだ救えていないみたいだ」 続きをみる
#LLMタグ
LLMに「right?」と聞くと同意が増える？ 45モデルの実験は、もっと...謎な結果だった
・先日、コードの設計についてLLMに相談していて、ふと引っかかりました。「〜でいいですよね？」と聞いた回だけ、心なしか賛成してもらえている気がしたんです。
・LLMに相談するとき、こちらの考えを先に言ってしまうことがあります。
cs.LG updates on arXiv.org
Local Violation Certification for Linear Predict-Then-Optimize Pipelines
・arXiv:2608.04474v1 Announce Type: new Abstract: Data-driven decision pipelines combining predictive machine learning models with downstream optimization software are increasingly used to make high-stakes operational decisions.
・Certifying the safety, fairness, and reliability of these decisions is essential, yet traditional scenario generation methods rely on repeated random testing, which becomes computationally proh
cs.LG updates on arXiv.org
Looking in the Mirror: Introspecting Side-Effect Misalignments Induced by Fine-Tuning
・arXiv:2608.04347v1 Announce Type: new Abstract: Fine-tuning enables a source model to acquire desired capabilities and behaviors in a target domain while retaining much of its general-purpose competence.
・However, this adaptation process can also degrade alignment properties that were present in the source model.
・Recent work has shown that large language models can be trained using LoRA-based modules known as introspe
Hugging Face Papers
Lossless Tensor Compression as Program Synthesis
Lossless Tensor Compression as Program Synthesis
cs.LG updates on arXiv.org
MALT: Lightweight Curvature-Aware Muon via Diagonal Preconditioning
・arXiv:2608.05088v1 Announce Type: new Abstract: Muon has recently emerged as a promising alternative to AdamW for language model pretraining by orthogonalizing momentum matrices using Newton-Schulz iterations.
・Although Muon mitigates gradient anisotropy, it does not explicitly account for the curvature geometry of the loss landscape and may therefore remain sensitive to curvature anisotropy.
・We bridge this gap by pro
cs.LG updates on arXiv.org
Manipulation-Proof Oblivious Audits against Deceptive Model Providers
・arXiv:2608.04365v1 Announce Type: new Abstract: Audits have emerged as a critical instrument for algorithmic governance, providing a mechanism for external scrutiny and governance of machine learning models.
・However, ensuring the integrity of such assessments remains a challenging issue.
・For instance in regulatory contexts, audits are typically declared or easily detected, thus enabling model providers to manipulate
cs.LG updates on arXiv.org
MarsCast: Transfer Learning of AI Weather Foundation Models to Planetary Atmospheres
・arXiv:2608.05054v1 Announce Type: cross Abstract: We investigate the transferability of Earth weather foundation models to planetary atmospheres by adapting the GraphCast graph neural weather forecasting model to Mars.
・While GraphCast achieves state-of-the-art performance for terrestrial forecasting, its applicability to non-Earth environments remains unexplored.
・Using the Mars Climate Database (MCD), which provides
cs.LG updates on arXiv.org
MemFly: On-the-Fly Memory Optimization via Information Bottleneck
・arXiv:2602.07885v2 Announce Type: replace-cross Abstract: Long-term memory enables large language model agents to tackle complex tasks through historical interactions.
・However, existing frameworks encounter a fundamental dilemma between compressing redundant information efficiently and maintaining precise retrieval for downstream tasks.
・To bridge this gap, we propose MemFly, a framework grounded in information bottle
cs.LG updates on arXiv.org
MemNovo: Look Back at the Spectrum for Balanced De Novo Peptide Sequencing from Mass Spectrometry
・arXiv:2606.11868v2 Announce Type: replace Abstract: De novo peptide sequencing from tandem mass spectrometry is pivotal in proteomics, enabling identification of novel peptides without reference databases.
・While recent Transformer-based encoder-decoder models have achieved remarkable performance, we uncover a critical pathology in their inference dynamics.
・Through comprehensive feature scaling experiments, we demonst
cs.LG updates on arXiv.org
MESH: Memory-Efficient Sinkhorn Optimization for Mixture-of-Experts Training
・arXiv:2608.04407v1 Announce Type: new Abstract: Memory-efficient matrix optimizers such as Sinkhorn gradient descent remove most AdamW optimizer state for dense Transformer matrices, but direct application to Mixture-of-Experts (MoE) training is unreliable.
・We study this failure in a controlled 110M-parameter nanowhale DeepSeek-style MoE pretraining setting.
・A SAGE/Sinkhorn hybrid reduces optimizer state from 0.883GB
WIRED
Meteor Showers, Eclipses, and More Are on the August 2026 Astronomical Calendar
・Here’s when some of this year’s most anticipated astronomical events will happen and the best way to observe them.
cs.LG updates on arXiv.org
MGSB: Manifold Gated Signature Branch Pressure-Domain Baseline Architecture for Two-Phase Pipeline Flows Under Distributional Shift
・arXiv:2608.04805v1 Announce Type: new Abstract: Leak detection models for multiphase pipelines often degrade when deployed under flow regimes that differ from training.
・Existing evaluations typically assess performance under in-distribution operating conditions, masking failures caused by regime transitions such as bubble-to-slug flow.
・We propose the Manifold Gated Signature Bias (MGSB), a regime-aware architecture c
WIRED
Microsoft’s Quantum Chief Doesn’t Care That Scientists Don’t Believe His Results
・Zulfi Alam thinks his team doesn’t need to “prove” they engineered a new state of matter in their bid to reinvent computing.
・Science would disagree.
機械学習タグが付けられた新着記事 - Qiita
Mii的な模式顔からのi2iで架空人物の顔多様性を上げたい
・はじめに 前回、同じデモグラフィックの架空科学者をプロンプトだけで描き分けましたが、ArcFace埋め込みの平均距離（InsightFace buffalo_1） による多様性スコアは最大 0.706 でした。
・そこで今回は、PILで顔幅、顎、目の大きさ・間隔などを変えた...
cs.LG updates on arXiv.org
Mind the Cap: Output-Budget Regimes Change the Measured Multilingual Reasoning Gap
・arXiv:2608.04160v1 Announce Type: cross Abstract: Multilingual evaluations report accuracy at a single output-token cap, but languages need different numbers of tokens to express the same content, so the cap is a hidden experimental variable.
・We test whether the native-vs-translate gap on MGSM (German, Thai, Swahili) is a token-budget artifact for Qwen3-8B and Llama-3.1-8B-Instruct under four prompting strategies.
cs.LG updates on arXiv.org
MINT: Tensor Decomposition on Stacked Recurrence Matrices for Time Series Data Mining
・arXiv:2608.04157v1 Announce Type: new Abstract: Recurrence plots are a time series data mining primitive applied to a variety of domains (e.g.
・star light curves, sound waveforms, CCT telemetry).
・This work proposes tensorized self-similarity matrices as a primitive for univariate time series datasets ($N\times n$) of $N$ time series of length $n$ with a subsequence window of length $m$, and whose tensor-based nature i
cs.LG updates on arXiv.org
MoCA: Multi-modal Cross-masked Autoencoder for Digital Health Measurements
・arXiv:2506.02260v4 Announce Type: replace-cross Abstract: Wearable devices enable continuous multi-modal physiological and behavioral monitoring, yet analysis of these data streams faces fundamental challenges including the lack of gold-standard labels and incomplete sensor data.
・While self-supervised learning approaches have shown promise for addressing these issues, existing multi-modal extensions present opportuni
cs.LG updates on arXiv.org
Modality Agreement- and Conflict-Aware Prototype Hypergraph Learning for Multimodal Intent Understanding
・arXiv:2608.04054v1 Announce Type: cross Abstract: Multimodal intent recognition requires understanding not only what textual, acoustic, and visual signals share, but also how they disagree.
・Such disagreement is frequently class-informative; for example, lexical positivity accompanied by incongruent vocal or facial behavior may indicate sarcasm or taunting, yet most fusion methods either encourage modality alignment o
cs.LG updates on arXiv.org
Model Inversion meets Cryptographic Fuzzy Extractors
・arXiv:2510.25687v4 Announce Type: replace-cross Abstract: Model inversion attacks pose an open challenge to privacy-sensitive applications that use machine learning (ML) models.
・For example, face authentication systems use modern ML models to compute embedding vectors from face images of the enrolled users and store them.
・If leaked, inversion attacks can accurately reconstruct user faces from the leaked vectors.
cs.LG updates on arXiv.org
MODEST: Multi-Optics Depth-of-Field Stereo Dataset
・arXiv:2511.20853v4 Announce Type: replace-cross Abstract: Training and evaluation of state-of-the-art computer vision algorithms for reliable shallow depth of field (DoF) rendering and defocus deblurring remain constrained by a persistent lack of large-scale, full-frame, high fidelity, real-image datasets.
・Optical effects of shallow DoF and defocus blur depend intimately on camera optical configuration set with focal
cs.LG updates on arXiv.org
Monsoon Mayhem to Market Waves: Forecasting Fisheries Resilience in Sri Lanka
・arXiv:2608.04023v1 Announce Type: cross Abstract: Sri Lanka's fisheries sector is important for jobs and food supply.
・Between 2019 and 2025, it faced several major problems at the same time, and how these events together affected fish production and prices is still not well understood.
・This study develops a framework to connect weather changes, major disruption events, fish production, and prices, with the goal of he
cs.LG updates on arXiv.org
MOON3.0: Reasoning-aware Multimodal Representation Learning for E-commerce Product Understanding
・arXiv:2604.00513v3 Announce Type: replace Abstract: With the rapid growth of e-commerce, exploring general representations rather than task-specific ones has attracted increasing attention.
・Although recent multimodal large language models (MLLMs) have driven significant progress in product understanding, they are typically employed as feature extractors that implicitly encode product information into global embedding
cs.LG updates on arXiv.org
Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting
・arXiv:2608.04455v1 Announce Type: cross Abstract: One of the most challenging problems entertainment live-streaming services face in recommendation systems is that user behaviors are sparse and delayed, and interaction data exhibits bias for different user segments.
・Unlike e-commerce applications where user actions follow linear sequences, live-streaming viewers engage in multiple concurrent behaviors of watching, ch
cs.LG updates on arXiv.org
Multicalibration Yields Better Matchings
・arXiv:2511.11413v2 Announce Type: replace Abstract: Consider the problem of finding the best matching in a weighted graph where we only have access to predictions of the actual stochastic weights, based on an underlying context.
・If the predictor is the Bayes optimal one, then computing the best matching based on the predicted weights is optimal.
・However, in practice, this perfect information scenario is not realistic
cs.LG updates on arXiv.org
Multimodal Alignment Through Joint Kernel Entropic Gromov--Wasserstein Optimal Transport
・arXiv:2608.04234v1 Announce Type: cross Abstract: We study the problem of aligning data from multiple modalities into a shared representation space, focusing on settings where strong pretrained unimodal encoders are available but cross-modal paired data are scarce.
・We propose a structure-preserving alignment framework, joint kernel entropic Gromov--Wasserstein Optimal Transport (JK-EGW), which maps multiple modalitie
cs.LG updates on arXiv.org
Multimodal Spatiotemporal Atmospheric Data Assimilation with Latent Flow-matching
・arXiv:2608.05103v1 Announce Type: new Abstract: Data assimilation (DA) uses Bayesian inference to update the state of a numerical forecast model with observed data.
・In this study, we propose a fundamentally different, unified approach to atmospheric data assimilation.
・We use latent video flow-matching to sample temporally consistent trajectories from a prior trained using ERA5 reanalysis (69 variables over an 8-day w
cs.LG updates on arXiv.org
MultiPathFormer: Towards a Foundation Model for Multipath Wireless Propagation
・arXiv:2608.05076v1 Announce Type: new Abstract: Recent advances in machine learning have enabled training of wireless foundation models, which aim to support tasks such as channel estimation, beam prediction, and localization based on wireless signals.
・Existing wireless foundation models typically pretrain on channel tensors using masked reconstruction over subcarriers, antennas, or time but ignore the physical chara
#LLMタグ
Muse Code、24時間の仕事を止めない「ログ」の正体
・21時48分、iTerm2 の下半分に赤い終了コードが出ていた。社内の検証用に、古い求人票の正規化を数百件まとめて回していた時だ。途中で止まるのは想定内で、たぶんリトライすれば済む。そう見ていた。
・でも、どの入力まで処理し、どの変換を通り、API 呼び出しを何回したのかを追い直すほうが面倒だった。ログはあるのに、仕事の経緯がない。「ここから再開」の一点が出せない。
AI News & Artificial Intelligence | TechCrunch
Naïve raises $28.5M to automate the grunt work of setting up and running a company
・Taking vibe-coding a step further, Naïve claims its infra can automate most of the work in setting up and running a business.
cs.LG updates on arXiv.org
Neighborhood-Aware Dual Biomedical Entity Linking
・arXiv:2608.04144v1 Announce Type: cross Abstract: Biomedical entity linking grounds mentions in clinical and scientific text to entities in a curated knowledge base (KB) with ontological structure, which supports downstream applications such as literature-scale information extraction and patient-record normalization.
・The task has several challenges at once: the KB contains large numbers of entities, mentions are ofte
cs.LG updates on arXiv.org
NeuMoSync: End-to-End Neuromodulatory Control for Plasticity and Adaptability in Continual Learning
・arXiv:2608.04358v1 Announce Type: cross Abstract: Continual learning (CL) requires models to learn tasks sequentially, yet deep neural networks often suffer from plasticity loss and poor knowledge transfer, which can impede their long-term adaptability.
・Drawing high-level inspiration from global neuromodulatory mechanisms in the brain, we introduce Neuromodulation and Synchronization (NeuMoSync), a novel architecture
cs.LG updates on arXiv.org
Neural Diversity Regularizes Hallucinations in Language Models
・arXiv:2510.20690v3 Announce Type: replace-cross Abstract: Language models continue to hallucinate despite increases in parameters, compute, and data.
・We propose neural diversity -- decorrelated parallel representations -- as a principled mechanism that reduces hallucination rates at fixed parameter and data budgets.
・While existing mitigation strategies largely target accuracy, we provide the first formal tail bounds
cs.LG updates on arXiv.org
NeuroPB: Scaling Neural Decoding with Pretrained Behavioral Representations
・arXiv:2608.04389v1 Announce Type: new Abstract: Decoding continuous motor trajectories from neural activity is essential for developing practical brain-computer interfaces (BCIs).
・However, current neural decoders are constrained by the limited scale and heterogeneity of neural recordings.
・In contrast, behavioral data can be collected more readily and at substantially larger scale from humans, animals, simulations, an
cs.LG updates on arXiv.org
NodeJEPA: Structure-Conditioned Latent Prediction for Node-Level Graph Self-Supervised Learning
・arXiv:2608.04381v1 Announce Type: new Abstract: Self-supervised learning on graphs is largely shaped by contrastive methods that depend on carefully designed augmentations, and by generative methods that reconstruct node attributes in the input space.
・Both paradigms can entangle representations with low-level input statistics rather than with relational structure.
・Joint-embedding predictive architectures (JEPA) inste
Hugging Face Papers
NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap
NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap
cs.LG updates on arXiv.org
Non-asymptotic implicit bias of logistic regression at early-stage gradient descent dynamics
・arXiv:2608.04382v1 Announce Type: new Abstract: Gradient descent has been of particular interest in modern machine learning beyond sole focus on optimization.
・Implicit bias emerging from optimization, though not being encoded by the learning objective, often prevents from overfitting to spurious patterns.
・A typical instance is the max-margin implicit bias of a linear classifier, widely established for exponentially t
cs.LG updates on arXiv.org
Non-Stationary Inventory Control with Lead Times
・arXiv:2602.05799v2 Announce Type: replace-cross Abstract: We study non-stationary single-item, periodic-review inventory control problems in which the demand distribution is unknown and may change over time.
・We analyze how demand non-stationarity affects learning performance across inventory models, including systems with demand backlogging or lost-sales, both with and without lead times.
・For each setting, we propose
cs.LG updates on arXiv.org
Nonparametric Goodness-of-fit Testing under Covariate Shift
・arXiv:2608.04860v1 Announce Type: cross Abstract: This paper develops procedures for nonparametric goodness-of-fit testing under covariate shift, where labelled data are drawn from a source population but goodness-of-fit is evaluated for a target population.
・The distribution mismatch is quantified by either a bounded moment condition or a sub-exponential tail condition on the target-to-source density ratio.
WIRED
Norton Coupon Codes: Up to 58% Off
・Whether you’re looking to protect your small business or your personal computer, we have the top coupons and deals to help you save at Norton.
cs.LG updates on arXiv.org
Not Every Divergence Should Be Suppressed: Counterfactual Recoverability in On-Policy Distillation
・arXiv:2608.04408v1 Announce Type: new Abstract: On-policy distillation (OPD) supervises student-visited trajectories, yet divergence-based rules cannot determine whether an erroneous prefix remains correctable.
・We formulate this decision as counterfactual recoverability and replay each error state through budget-matched teacher-continuation and rollback branches.
・Based on their relative success, states are categorize
cs.LG updates on arXiv.org
NuclearDiffusion: Text-to-Image Foundation Models for Learning Nuclear Energy Concepts
・arXiv:2608.04030v1 Announce Type: cross Abstract: Generative artificial intelligence (AI) has transformed text-to-image synthesis, yet its ability to represent specialized engineering domains remains largely unexplored.
・As an exmaple in nuclear engineering, general-purpose foundation models frequently generate physically incorrect or conceptually inconsistent images because they lack domain-specific knowledge.
WIRED
NZXT Discount Codes: 50% Off in August 2026
・Save 50%, plus up to $250 with NZXT promo codes and discounts.
cs.LG updates on arXiv.org
OctoLong: Mid-Training On Cross-Repository Code Contexts Enhances Long-Context Modeling
・arXiv:2608.05141v1 Announce Type: cross Abstract: Context lengths of language models (LMs) have dramatically increased, driven by the demands for in-context learning, self-improvement, and long-horizon agentic workflows.
・Existing long-context corpora, however, are dominated by books, academic articles, and code repositories, which are finite resources and often scarce in long-distance dependencies.
・In this work, we i
cs.LG updates on arXiv.org
ODRA: Synthesizing Cognitive Behavioral Therapy Sessions with Structured Chain-Of-Thought and Dynamic Patient Resistance
・arXiv:2608.04524v1 Announce Type: cross Abstract: Synthetic generation of Cognitive Behavioral Therapy (CBT) sessions is challenged by two competing demands: adhering to strict therapeutic structure while modeling the resistant, unpredictable behavior of real patients.
・Existing script-based methods fail to capture dynamic therapeutic interactions, while multi-agent approaches struggle to adhere to CBT's sequential st
#LLMタグ
Office文書を数ミリ秒で「LLMが読めるMarkdown」に変換するOSS「anydoc」とは
・Word、PowerPoint、Excel、PDFなどの社内文書をAIに読ませようとしたとき、意外に大きな問題になるのが「文書をどうテキスト化するか」です。
・単純に文字だけを抜き出すと、見出し、表、箇条書き、脚注といった文書構造が崩れてしまいます。反対に、高機能な文書解析ツールは、処理が重かったり、外部サービスへのファイル送信が必要だったりします。
AI News & Artificial Intelligence | TechCrunch
Omilia raises $67M to scale its customer support platform
・The Series B is the company's second fundraise since it last raised capital in 2020.
・In that time, it has increased its ARR by 10x to $60 million.
cs.LG updates on arXiv.org
On Hamming-Lipschitz Type Stability of the Subdominant (Minmax) Ultrametric: Theory and Simple Proofs
・arXiv:2608.04014v1 Announce Type: new Abstract: The subdominant (minmax) ultrametric is a canonical tree-structured summary of a dissimilarity matrix, arising equivalently as the ultrametric induced by single-linkage clustering.
・While its classical stability theory is usually formulated in $\ell_\infty$ or Gromov--Hausdorff terms, such bounds are poorly suited to sparse perturbations that alter only a few pairwise di
cs.LG updates on arXiv.org
On MUON optimization: From non-convergence to an error analysis with Polar Express and the Newton-Schulz polynomial from implementations
・arXiv:2608.04607v1 Announce Type: cross Abstract: Stochastic gradient descent (SGD) optimization methods are the standard instruments for the training of deep neural networks (DNNs).
・In many relevant artificial intelligence (AI) systems - such as popular large language models (LLMs)-not the standard SGD scheme is used as the optimization method but instead suitable accelerated variants of SGD are employed.
cs.LG updates on arXiv.org
One Surrogate to Fool Them All: Universal, Transferable, and Targeted Adversarial Attacks with CLIP
・arXiv:2505.19840v3 Announce Type: replace-cross Abstract: Deep Neural Networks (DNNs) have achieved widespread success yet remain prone to adversarial attacks.
・Typically, such attacks either involve frequent queries to the target model or rely on surrogate models closely mirroring the target model -- often trained with subsets of the target model's training data -- to achieve high attack success rates through transfe
Hugging Face Papers
OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents
OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents
cs.LG updates on arXiv.org
OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents
・arXiv:2608.05013v1 Announce Type: cross Abstract: LLM agents are increasingly applied to open-ended everyday requests that span work, study, and life.
・These tasks are long-horizon, cross-environment, and multimodal, forcing the agent to preserve goals and constraints across many steps while navigating heterogeneous tools and attachments.
・While prior work has addressed individual failure modes such as goals drift, sta
Hugging Face Papers
OPD-V: Visual On-Policy Self-Distillation with Modality Balance
OPD-V: Visual On-Policy Self-Distillation with Modality Balance
The Verge
OpenAI is giving ChatGPT free users unlimited text chats
・OpenAI is making a big change for ChatGPT users on its free and Go tiers: starting next week, users on those tiers will be able to have unlimited text chats with the chatbot, according to OpenAI.
・Right now, you may run into rate limits if you do too many text chats on those tiers, but those limits, just for text, will be going away soon.
・Messages that include things like file uploads and images will continue to have
AI News & Artificial Intelligence | TechCrunch
OpenAI says Apple’s own security practices undermine its trade secrets case
・Newly filed court exhibits show OpenAI’s legal strategy in Apple’s trade secrets lawsuit: argue that Apple’s own security and offboarding practices — including allowing an Apple manager to access a former engineer’s iCloud account after he left the company —undermine its claims that the allegedly stolen information was properly protected.
cs.LG updates on arXiv.org
Optimal Training-Time Scaling in Gradual Adaptation
・arXiv:2608.04927v1 Announce Type: new Abstract: In gradual adaptation, how should the training time on each task change as the number of intermediate tasks increases?
・We study this question for overparameterized linear regression tasks that change smoothly and share a zero-loss solution.
・With $N$ tasks and training time $s_N$ on each, the final learning progress converges to a continuum curve when $Ns_N\to\tau$.
cs.LG updates on arXiv.org
Optimizing What Policies Learn From: Recoverability-aware Rollout Intervention Learning
・arXiv:2608.05080v1 Announce Type: new Abstract: Critic-free group-based reinforcement learning has become a scalable approach for post-training large language models.
・However, most existing methods allocate the same number of rollouts to every task and trajectory state, even though some rollouts provide much more useful learning signals than others.
・Recent work has started to treat rollout generation as an adaptive d
cs.LG updates on arXiv.org
Out-Of-The-Loop Multi-Fidelity Bayesian Optimization
・arXiv:2608.04113v1 Announce Type: new Abstract: Black-box optimization is a ubiquitous problem in science and engineering, often dealing with expensive objective functions with cheaper lower-fidelity proxies available.
・Multi-fidelity Bayesian optimization (MF-BO) is a principled approach to this problem, leveraging correlations across different fidelities when querying the objective.
・However, for many important MF-BO
cs.LG updates on arXiv.org
Personalized Federated Sparse Adaptation of Time-Series Foundation Models
・arXiv:2608.04695v1 Announce Type: new Abstract: Federated adaptation of time-series foundation models (TSFMs) is attractive for building energy forecasting because meter data are private, distributed, and highly non-IID.
・However, a single parameter-sharing strategy is unlikely to serve all pretrained TSFMs or building clients: fully shared adapters can suppress building-specific temporal behavior, while fully local a
cs.LG updates on arXiv.org
Physics-informed reduced-order modelling with equivariant spectral submanifolds
・arXiv:2608.04239v1 Announce Type: new Abstract: Spectral submanifold (SSM) reduction has emerged as a mathematically principled route to reliable nonlinear reduced-order models, capturing dynamics beyond the reach of linear techniques such as Dynamic Mode Decomposition (DMD).
・The computation of SSMs, however, remains computationally expensive, particularly for high-dimensional systems.
・In this work, we introduce equi
Hugging Face Papers
Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models
Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models
cs.LG updates on arXiv.org
Predicting Brain Morphometry with MT-GNN: Mesh Evolution in Continuous Time with Graph-Based Metric Tensor Embeddings
・arXiv:2608.05132v1 Announce Type: cross Abstract: Predicting how a subcortical structure's shape will evolve from a few prior scans could support prognosis and clinical-trial enrichment.
・Existing longitudinal mesh predictors either extrapolate shape trajectories via high-dimensional embeddings or regress vertex deformations directly.
・We instead predict the surface's intrinsic geometry in continuous time: a single per
cs.LG updates on arXiv.org
PriDyG: Privacy-preserving Dynamic Graph Inference with LLM-GNN Collaboration
・arXiv:2608.04255v1 Announce Type: cross Abstract: Graph inference over relational data can expose sensitive edge information, and this risk becomes more severe in dynamic graphs, where repeated model updates cause privacy loss to accumulate.
・We formulate Edge-level Differentially Private Dynamic Graph Inference (EDG) and propose PriDyG, a private inference framework that combines GNN-based structural learning with LL
MarkTechPost
Prime Intellect Releases Prime Agent: An Open-Source RLM Harness Where Sub-Agents Are Function Calls Inside Persistent IPython Kernel
・Prime Intellect has open-sourced Prime Agent, a coding and research harness built on two abstractions: the Recursive Language Model, which turns sub-agent calls into functions inside a persistent IPython kernel, and the Continual Harness, which lets the agent edit its own prompts, skills, memory, and sub-agent specs mid-run.
・With Opus 5 it reports 95.5% RHAE Best@1 on ARC-AGI-3, above the reported human expert baseli
cs.LG updates on arXiv.org
Privileged, but Biased: How PI-Conditioned Teachers Break Self-Distillation
・arXiv:2608.04794v1 Announce Type: cross Abstract: Self-distillation (SD) has emerged as a compute-efficient alternative to reinforcement learning with verifiable rewards: a self-teacher, conditioned on privileged information (PI) about the answer such as a reference solution, supplies dense per-token supervision to a student that never sees it.
・Reported gains, however, come almost exclusively from narrow, low-difficu
cs.LG updates on arXiv.org
Protoreasoning in Tiny Transformers
・arXiv:2608.04980v1 Announce Type: cross Abstract: We show that tiny transformers can profitably employ a simple form of Chain of Thought, which we call protoreasoning, allowing us to study step-by-step reasoning on ~1M-parameter models and opening up opportunities for much more detailed experimentation and analysis than is feasible for larger models.
・Current Large Language Models exhibit impressive step-by-step reaso
cs.LG updates on arXiv.org
Prototype-based Self-Supervised Multimodal Learning for PPG and Accelerometry Signals
・arXiv:2510.09764v2 Announce Type: replace Abstract: Modeling multi-modal time-series data is critical for capturing system-level dynamics, particularly in biosignals where modalities such as ECG, PPG, EDA, and accelerometry provide complementary perspectives on interconnected physiological processes.
・While recent self-supervised learning (SSL) advances have improved unimodal representation learning, existing multi-mo
cs.LG updates on arXiv.org
Provable Limits and Certified Deferral for Verbalized Uncertainty in Small Language Models
・arXiv:2608.05064v1 Announce Type: cross Abstract: Small open-weight language models increasingly run in private, offline, and cost-sensitive settings, where the key deployment question is not only what a model answers but when it should defer to a human.
・We study whether verbalized confidence can support risk-controlled deferral, evaluating eleven instruction-tuned models from three families, 0.5B to 14B parameters,
cs.LG updates on arXiv.org
PSI3D: Plug-and-Play 3D Stochastic Inference with Slice-wise Latent Diffusion Prior
・arXiv:2512.18367v2 Announce Type: replace-cross Abstract: Diffusion models are highly expressive image priors for Bayesian inverse problems.
・However, most diffusion models cannot operate on large-scale, high-dimensional data due to high training and inference costs.
・In this work, we introduce a Plug-and-play algorithm for 3D stochastic inference with latent diffusion prior (PSI3D) to address massive ($1024\times 1024
cs.LG updates on arXiv.org
Pun Intended: Multi-Agent Translation of Wordplay with Contrastive Learning and Phonetic-Semantic Embeddings for CLEF JOKER 2025 Task 2
・arXiv:2507.06506v2 Announce Type: replace-cross Abstract: Translating wordplay across languages presents unique challenges that have long confounded both professional human translators and machine translation systems.
・This research proposes a novel approach for translating puns from English to French by combining state-of-the-art large language models with specialized techniques for wordplay generation.
・Our methodolo
cs.LG updates on arXiv.org
Random features for Grassmannian kernel approximation with bounded rank-one projections
・arXiv:2608.04227v1 Announce Type: new Abstract: We propose a family of random feature maps for scalable kernel machines on low-dimensional subspaces, ie on the Grassmannian manifold.
・Such representations are useful when data classes or clusters are well described by the span of a few samples.
・Classical Grassmannian kernels, including the projection and Binet-Cauchy kernels, require full Gram matrices, which leads to
cs.LG updates on arXiv.org
Real-time probabilistic tsunami forecasting via generative AI
・arXiv:2608.04327v1 Announce Type: new Abstract: Explicit onshore tsunami inundation forecasting can improve public risk awareness, but deterministically predicted inundation boundaries under highly uncertain conditions, such as near-field tsunamis generated by megathrust earthquakes, may falsely imply safety outside the boundaries.
・Consequently, current warnings primarily target coastal tsunami height, not onshore in
cs.LG updates on arXiv.org
Recurrent Residual Quantization: A Progressive Multi-Precision Representation for LLMs
・arXiv:2608.04048v1 Announce Type: new Abstract: Serving large language models (LLMs) under diverse deployment constraints requires flexible trade-offs between accuracy, memory footprint, and throughput.
・However, conventional quantization methods typically require a separate checkpoint for each target bit-width.
・We introduce Recurrent Residual Quantization (RRQ), a post-training quantization (PTQ) framework that repre
cs.LG updates on arXiv.org
Regularization can make diffusion models more efficient
・arXiv:2502.09151v3 Announce Type: replace Abstract: Diffusion models are one of the key architectures of generative AI.
・Their main drawback, however, is the computational costs.
・This study indicates that the concept of sparsity, well known especially in statistics, can provide a pathway to more efficient diffusion pipelines.
cs.LG updates on arXiv.org
Reinforcement Learning and Consumption-Savings Behavior
・arXiv:2510.20748v2 Announce Type: replace-cross Abstract: This paper demonstrates how reinforcement learning can explain two puzzling empirical patterns in household consumption behavior during economic downturns.
・I develop a model where agents use Q-learning with neural network approximation to make consumption-savings decisions under income uncertainty, departing from standard rational expectations assumptions.
cs.LG updates on arXiv.org
Relational Response Fields: A General Theory of Black-Box LLM Response Consistency and Recovery
・arXiv:2608.04552v1 Announce Type: cross Abstract: Black-box language-model reliability is commonly pursued by sampling, prompting, voting, verifying, or iteratively revising individual answers.
・We ask a prior question: \emph{what determines whether a collection of black-box responses is recoverable at all?} We represent responses to typed transformations of a query as a \emph{relational response field} (RRF).
cs.LG updates on arXiv.org
Relevant but Incomplete: Referential Dangling as a Paradigm-Level Failure Mode in Hard Prompt Compression
・arXiv:2608.04569v1 Announce Type: cross Abstract: Hard prompt compression reduces long-context inference cost by independently scoring tokens, sentences, or chunks and retaining the highest-scoring units under a budget.
・We identify a structural failure in this procedure: independent selection can split dependent evidence pairs, retaining one member while deleting the other.
・When retained text contains an answer but d
cs.LG updates on arXiv.org
Representational separation between unitary and channel quantum generative models via shared classical randomness at shallow depth
・arXiv:2608.05110v1 Announce Type: cross Abstract: Near-term quantum hardware limits circuit depth and often imposes geometrically local connectivity for quantum generative models, restricting the output distributions accessible to shallow unitary Born models.
・Introducing stochasticity into a unitary quantum Born model can improve the empirical generative performance of the resulting channel model and, for a restricte
cs.LG updates on arXiv.org
RESample: A Robust Data Augmentation Framework via Exploratory Sampling for Robotic Manipulation
・arXiv:2510.17640v4 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models have shown strong manipulation capability when trained with large-scale imitation learning datasets.
・However, these datasets that predominantly consist of successful trajectories rarely provide the corrective supervision required when execution deviates from standard demonstrations.
・During deployment, these physical deviatio
cs.LG updates on arXiv.org
Rethinking Reservoir Pruning: A Dynamical Perspective for Echo State Networks
・arXiv:2608.04593v1 Announce Type: new Abstract: Echo State Networks (ESNs) offer an efficient framework for temporal prediction, but their randomly initialized reservoirs are often over-parameterized and dynamically redundant.
・Existing pruning methods largely rely on static connectivity or activation statistics, which may overlook neurons that shape input-driven state transitions.
・We propose Dynamical Mode Pruning (D
cs.LG updates on arXiv.org
Revealed Rationality: Label-Free Evaluation and Regularization from Representation Theorems
・arXiv:2608.05015v1 Announce Type: cross Abstract: Representation theorems in decision theory establish that behavior satisfies certain axioms if and only if it can be rationalized by a well-defined objective.
・I argue that this ``if and only if'' structure provides a potentially useful foundation for label-free evaluation and regularization of LLMs and other AI systems.
・Axiom compliance can be checked from the model's
cs.LG updates on arXiv.org
Review Text as a Leading Indicator of Displayed Reputation in Platform Rating Systems: Evidence from 34 U.S. Short-Term Rental Markets
・arXiv:2504.14053v2 Announce Type: replace-cross Abstract: Rating systems on accommodation platforms suffer from a familiar problem: nearly every listing displays a nearly perfect score, so the number that is supposed to separate good listings from bad ones barely varies.
・Whether the review text accumulating beneath those scores still carries usable information is an open question.
・I ask a dynamic version of it: does
cs.LG updates on arXiv.org
Reward Structure Shapes the Interaction Between Episodic Exploration and Neural Memory in Reinforcement Learning
・arXiv:2608.05111v1 Announce Type: new Abstract: In partially observable reinforcement learning, agents face a dual bottleneck: they must explore to encounter rewarding states and retain that experience in memory to optimize their policies.
・Exploration bonuses and memory architectures are traditionally evaluated in isolation, leaving their interaction unmeasured, and standard notions of sparse reward conflate temporal
cs.LG updates on arXiv.org
RiboSphere: Learning Unified and Efficient Representations of RNA Structures
・arXiv:2603.19636v2 Announce Type: replace Abstract: Accurate RNA structure modeling remains difficult because RNA backbones are highly flexible, non-canonical interactions are prevalent, and experimentally determined 3D structures are comparatively scarce.
・We introduce RiboSphere, a framework that learns discrete geometric representations of RNA by combining vector quantization with flow matching.
・Our design is motiv
cs.LG updates on arXiv.org
Right Reset: Chunking by Prefix Removal
・arXiv:2608.04330v1 Announce Type: cross Abstract: Removing the left context from a causal language model reveals a useful kind of boundary: an edge where the model processes the same right-hand tokens with little change.
・We turn this observation into prefix-removal probing and introduce Right Reset (RR), which measures preservation of the right-hand hidden-state trajectory.
・A dynamic program converts RR edge scores i
cs.LG updates on arXiv.org
RingSQL: Schema-Independent Synthetic Data Generation for Text-to-SQL Reinforcement Learning
・arXiv:2601.05451v2 Announce Type: replace Abstract: Recent advances in text-to-SQL have been driven by larger models, better datasets, and new training methods like RLVR.
・However, progress remains limited by scarce high-quality training data, a problem RLVR is especially sensitive to since noisy data can produce spurious rewards.
・Manual data creation is expensive, and existing synthetic methods trade off reliability
cs.LG updates on arXiv.org
Robust and Efficient Motion Reasoning for Privacy-Aware Classroom Incident Recognition
・arXiv:2608.05115v1 Announce Type: cross Abstract: Can computer vision help make classrooms safer?
・In this pilot study, we investigate privacy-aware and computationally efficient classroom incident recognition from CCTV-style observations.
・This setting remains underexplored, with limited benchmarks and few methods designed for the privacy, efficiency, and generalization demands of real-world deployment.
cs.LG updates on arXiv.org
Robust and Personalized Federated Learning for Aircraft-Engine Prognostics under Benign and Adversarial Client Heterogeneity
・arXiv:2608.04045v1 Announce Type: new Abstract: Federated learning (FL) enables aircraft fleet operators to jointly train remaining-useful-life (RUL) models from engine sensor telemetry without sharing raw data.
・This study examines two complementary challenges: benign heterogeneity, where honest operators observe different operating conditions and fault modes, and adversarial heterogeneity, where compromised operator
cs.LG updates on arXiv.org
Robust Control under Stationary Ambiguity
・arXiv:2608.04832v1 Announce Type: new Abstract: Control policies optimized in simulation can perform poorly in the real system when the parameters $x$ of the simulator are estimated from limited data but the resulting parameter uncertainty is not represented inside the simulation.
・A common way to incorporate such ambiguity is to simulate each trajectory of the system under a randomly drawn value for $x$.
・Since the po
cs.LG updates on arXiv.org
Robustness Emerges Early in Training Dynamics, but Is Not Preserved
・arXiv:2608.04442v1 Announce Type: new Abstract: Robustness to natural corruptions remains a fundamental challenge for deep neural networks.
・In this paper, we identify a robustness fading phenomenon where shallow layers spontaneously develop robust representations and flat loss landscapes in early training, yet these properties are not preserved during standard convergence.
・To address this, we propose a framework that
cs.LG updates on arXiv.org
RooflineBench: A Benchmarking Framework for On-Device LLMs via Roofline Analysis
・arXiv:2602.11506v4 Announce Type: replace Abstract: The transition toward localized intelligence through Small Language Models (SLMs) has intensified the need for rigorous performance characterization on resource-constrained edge hardware.
・However, objectively measuring the theoretical performance ceilings of diverse architectures across heterogeneous platforms remains a formidable challenge.
・In this work, we propose
cs.LG updates on arXiv.org
Sample Complexity of Multicalibration for Multilevel Properties
・arXiv:2608.04288v1 Announce Type: new Abstract: Calibration requires a predictor to be unbiased after conditioning on its own predictions.
・Multicalibration asks for this guarantee simultaneously across a collection of groups.
・Many prediction tasks ask for several related features of the same conditional outcome distribution: variance is defined relative to the mean, skewness relative to both mean and variance, and co
WIRED
Samsung Odyssey G8 6K Review: Pixel Overdose
・Samsung dared to bring 6K to the world of gaming monitors before our PCs were truly ready for it.
WIRED
Sealy Promo Codes: $100 Off
・Whether you’re switching from springs to memory foam or just want to sleep cooler this summer, these Sealy mattress deals will have you waking up refreshed—and with cash left in your wallet.
WIRED
Shark PowerDetect Speed Review (2026): Light and Powerful
・Shark’s new PowerDetect Speed is lighter and more affordable than the original, but still managed to ace most of my vacuum tests.
cs.LG updates on arXiv.org
Short-term load forecasting under EU-AI Act Requirements in Safety-Critical Environments: Results from a 41-day live challenge on the aggregated German transmission-grid load
・arXiv:2608.05018v1 Announce Type: cross Abstract: Short-term load forecasting (STLF) play a vital role in the electric power industry.
・It serves infrastructure that European and German law designate as critical.
・Determinism, reproducibility, and auditability are engineering requirements rather than optional extras.
Hugging Face Papers
SIGNPOST-Bench: Benchmarking Text-Vision Conflict Resolution in Multimodal Large Language Models
SIGNPOST-Bench: Benchmarking Text-Vision Conflict Resolution in Multimodal Large Language Models
cs.LG updates on arXiv.org
SiMDex: Mining Similar Egocentric Videos for Cross-Embodiment Dexterous Manipulation
・arXiv:2608.04196v1 Announce Type: cross Abstract: Recent years have witnessed an explosive trend of scaling ego-centric human videos for robot manipulation, yet it remains unclear which data actually benefits dexterous manipulation.
・We present SiMDex, a similarity-based data mining framework that casts human data selection for VLA post-training in dexterous manipulation as a recommendation problem.
・For each robot dem
cs.LG updates on arXiv.org
Simultaneous estimation of multiple discrete unimodal distributions under stochastic order constraints
・arXiv:2603.11532v2 Announce Type: replace-cross Abstract: We study the problem of estimating multiple discrete unimodal distributions, motivated by search behavior analysis on a real-world platform.
・To incorporate prior knowledge of precedence relations among distributions, we impose stochastic order constraints and formulate the estimation task as a mixed-integer convex quadratic optimization problem.
・Experiments on
cs.LG updates on arXiv.org
SJEPA: Learning Elegant Latent Dynamics with Hybrid Symbolic-Neural Predictors
・arXiv:2608.04060v1 Announce Type: new Abstract: Joint-embedding predictive architectures learn abstract states by predicting target embeddings from context embeddings, but their transition models are typically opaque neural maps.
・We introduce SJEPA, a reconstruction-free JEPA framework that learns predictive representations whose induced dynamics admit compact symbolic descriptions.
・Its hybrid transition combines a s
Hugging Face Papers
SKILL-KD: Contrastive Skill Distillation for LLM Agents
SKILL-KD: Contrastive Skill Distillation for LLM Agents
The Verge
SoftBank donated $50 million to Trump’s library months before federal data center deal
・The Portsmouth, Ohio site where SoftBank will build its data center.
・| Image: US Department of Energy SoftBank contributed $50 million to the Trump Presidential Library in January, just months before announcing that it's leasing land from the federal government to build a sprawling data center in Ohio.
・The Japanese company revealed the timing in response to a June letter from Sen.
ITmedia NEWS 最新記事一覧
SpaceXのロケット上段、月面に衝突 NASAが発表 「地球や月面機器に危険なし」
・NASAは、SpaceXのFalcon 9使用済み上段が月面に衝突したと発表した。計画外の衝突で、地球や月面機器への危険はない。SpaceXは、月への高エネルギーミッションでは通常の制御廃棄が難しい場合があると説明した。
cs.LG updates on arXiv.org
SparseDitto: Customizing GPU Kernels for Different Sparsity Patterns with LLM-Based Agentic System
・arXiv:2608.05033v1 Announce Type: cross Abstract: Sparse matrix kernels are fundamental to scientific computing, graph analytics, and machine learning.
・Their GPU performance depends strongly on the input sparsity pattern and execution strategy.
・For the same SpMM on the same matrix, cuSPARSE exhibits a 350x performance gap between CSR and Blocked-ELL.
cs.LG updates on arXiv.org
Spatiotemporal Graph Transformer for Traffic Intelligence in Edge Computing
・arXiv:2608.04075v1 Announce Type: new Abstract: Accurate traffic forecasting is essential for proactive resource management in edge computing, where service demand evolves dynamically across both space and time.
・In practical cellular edge systems, traffic exhibits strong spatial correlations among neighboring service regions and long-range temporal dependencies driven by user mobility and application behavior.
cs.LG updates on arXiv.org
SpecDrop: Parameter-Free Category-Conditioned Routing for Modular Specialization
・arXiv:2608.04084v1 Announce Type: new Abstract: Mixture-of-experts (MoE) networks pursue specialization through learned routers, gates, and load-balancing losses, yet at matched total-parameter budgets learned routers can underperform equal-weight No-Routing baselines.
・Is the bottleneck the routing algorithm, or the alignment between training-signal granularity and the target categories?
・We probe the question with Sp
cs.LG updates on arXiv.org
SpecRoll: Fast-Slow Verifier-Feedback Adaptation for Speculative Reinforcement Learning Rollouts
・arXiv:2608.04962v1 Announce Type: new Abstract: Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but autoregressive rollout generation remains a major efficiency bottleneck.
・Speculative decoding can accelerate generation, yet applying it during RL is difficult because the target policy continually evolves: static proposers become stale, while frequent drafter upd
cs.LG updates on arXiv.org
Spend Bits Where Queries Look: KV Cache Vector Quantization with Attention-Preserving Transforms
・arXiv:2608.04074v1 Announce Type: new Abstract: Long-context LLM decoding reads the key-value (KV) cache at every step.
・Loading it takes longer than computing attention over it, so throughput is bandwidth-bound.
・Hence, reducing the cache size can raise both decoding speed and serving capacity.
cs.LG updates on arXiv.org
SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation
・arXiv:2608.04419v1 Announce Type: new Abstract: On-policy distillation (OPD) provides dense teacher supervision on student-generated trajectories, but standard reverse-KL training can assign insufficient probability to other plausible continuations.
・Teacher entropy alone does not reveal whether uncertainty is concentrated among a few plausible next tokens or dispersed over a long probability tail, nor whether the stu
cs.LG updates on arXiv.org
SSTQ:Privacy-Preserving Vector Quantization via Subsampled Stochastic TurboQuant
・arXiv:2608.05127v1 Announce Type: new Abstract: Achieving local differential privacy in distributed optimization while maintaining low communication cost remains challenging.
・Existing vector quantization methods, such as vqSGD, use high-dimensional geometric constructions but incur unfavorable dimension-dependent variance.
・In this work, we propose Subsampled Stochastic TurboQuant (SSTQ), a framework that combines ove
cs.LG updates on arXiv.org
Stabilizing Multi-Attack Adversarial Training via Bandit Optimization
・arXiv:2511.12265v2 Announce Type: replace Abstract: Deep Neural Networks (DNNs) remain vulnerable to diverse adversarial perturbations, motivating multi-attack adversarial training (AT) for improved robustness.
・However, existing methods either incur prohibitive overhead by computing all attacks at each iteration, or rely on stochastic sampling over adversarial examples, which may cause excessive parameter drift.
cs.LG updates on arXiv.org
Stable Density Ridges: Consistency and Convergence of Subspace Constrained Mean Shift
・arXiv:2608.05112v1 Announce Type: cross Abstract: The Subspace Constrained Mean Shift (SCMS) algorithm is a popular nonparametric method for extracting density ridges, which serve as a low-dimensional representation of high-dimensional data.
・It is a widely held belief in the literature that SCMS trajectories converge to the classical density ridge, which we call the "static ridge", defined via the density gradient an
cs.LG updates on arXiv.org
Stable GFlowNets with TV Monitoring and Probabilistic Guarantees
・arXiv:2605.01729v2 Announce Type: replace Abstract: Generative Flow Networks (GFlowNets) learn to sample states proportional to an unnormalized reward.
・Despite their theoretical promise, practical training is often unstable, exhibiting severe loss spikes and mode collapse.
・To tackle this, we first assess the sensitivity of GFlowNet objectives, demonstrating that a small Total Variation (TV) distance between the learn
cs.LG updates on arXiv.org
State2State: Environment-Derived Mid-Training for LLM Agents
・arXiv:2608.04934v1 Announce Type: cross Abstract: Training LLM agents commonly relies on supervised fine-tuning from expert trajectories or online reinforcement learning over human-specified tasks with handcrafted verifiers.
・Though effective, both remain bottlenecked by externally specified tasks and supervision signals, limiting the scalability and diversity of agent training.
・We study an environment learning paradi
cs.LG updates on arXiv.org
Statistical learning theory and Occam's razor: Regularization
・arXiv:2608.04049v1 Announce Type: cross Abstract: The principle of Occam's razor, which instructs us to prefer simplicity in inductive inference, has attracted much scrutiny both in the philosophy of science and in machine learning.
・In either field, however, a justification for the principle has been elusive.
・In this paper, building on an earlier "core argument," I spell out a justification from statistical learning
WIRED
Stearns and Foster Promo Codes: $300 Off in August 2026
・Discover the best Stearns and Foster promo codes, discounts, and deals to elevate your sleep experience with premium comfort and quality.
cs.LG updates on arXiv.org
Stochastic Emulation using Generalized Stratified Sampling for Performance-Based Risk Optimization of Structures
・arXiv:2608.05006v1 Announce Type: new Abstract: Metamodels are instrumental in reducing the computational burden associated with nested reliability analyses and optimization loops in Performance-Based Risk Optimization (PBRO) of structures under stochastic loads.
・In this context, stochastic emulators are particularly useful because they approximate response distributions while accounting for the intrinsic stochastici
cs.LG updates on arXiv.org
stratum: A System Infrastructure for Massive Agent-Centric ML Workloads
・arXiv:2603.03589v3 Announce Type: replace-cross Abstract: Recent advances in large language models (LLMs) transform how machine learning (ML) pipelines are developed and evaluated.
・LLMs enable a new type of workload, agentic pipeline search, in which autonomous or semi-autonomous agents generate, validate, and optimize complete ML pipelines.
・These agents predominantly operate over popular Python ML libraries and exhi
cs.LG updates on arXiv.org
Sublogarithmic Swap Regret in Multiplayer General-Sum Games via Hybrid Regularization
・arXiv:2608.04149v1 Announce Type: cross Abstract: Swap regret governs the rate at which uncoupled learning dynamics converge to correlated equilibria in multiplayer general-sum games.
・Under full-information feedback, the best previous guarantee when every player follows the same dynamics grows logarithmically in the horizon $T$.
・We construct uncoupled dynamics under which every player incurs only $O(nm^2\sqrt{\log m\
cs.LG updates on arXiv.org
Suppression Sticks, Locality Is Fragile: A Closed-Loop Target-and-Control Audit of Task-Vector Negation in VLA Policies
・arXiv:2608.04692v1 Announce Type: cross Abstract: Task-vector arithmetic offers a closed-form way to modify a model, yet its behavioral locality remains unclear in closed-loop robot control.
・We present a target-and-control audit of per-skill task-vector subtraction from multitask vision-language-action (VLA) policies.
・Across all ten LIBERO-Goal skills, subtraction produces three qualitatively different regimes: targe
cs.LG updates on arXiv.org
SVI-DAG: A Structured Variational Inference Approach to Bayesian Causal Discovery
・arXiv:2608.04930v1 Announce Type: new Abstract: Bayesian causal discovery seeks to determine the posterior distribution of causal theories, which are interpreted as directed acyclic graphs (DAGs) that explain the observed data.
・The resulting posterior allows systematic reasoning regarding epistemic uncertainty within these theories.
・Nonetheless, finding such graphs is difficult due to identifiability problems and lim
cs.LG updates on arXiv.org
Tactus: Open-Vocabulary Object Recognition from Low-Cost Pressure Arrays
・arXiv:2608.04043v1 Announce Type: new Abstract: Resistive pressure arrays are the cheapest and most widely shipped tactile sensors, yet tactile representation learning has concentrated on optical sensors that image a deforming gel.
・We present Tactus, an open model that answers text queries from pressure data alone: on the STAG benchmark (27 objects, held-out recordings), it reaches 0.771 +/- 0.062 top-1 over four run
cs.LG updates on arXiv.org
The Cost of Binarizing Survival Outcomes in Clinical Prognostic Modeling
・arXiv:2608.04046v1 Announce Type: cross Abstract: Survival analysis is an established framework for analyzing time-to-event data, yet many clinical machine learning studies still binarize the outcome before model training.
・This practice excludes censored patients, collapses temporal information into a single threshold, and can affect which features are selected as prognostically relevant.
・We examine the cost of this
cs.LG updates on arXiv.org
The Hamilton-Jacobi Theory of Deep Learning
・arXiv:2605.28983v2 Announce Type: replace Abstract: In this paper, training a neural network is identified, exactly, as a search through Hamilton--Jacobi initial-value problems: each gradient step selects the initial data of a viscous Hamilton--Jacobi equation whose Hopf--Cole propagator best fits the observations; at inference, the input is the spatial point at which that solution is evaluated and the initial condit
cs.LG updates on arXiv.org
The Loss Does Not See the Basis, but Adam Does
・arXiv:2608.05136v1 Announce Type: new Abstract: Gradient descent on a factored model $W = UV^\top$ is implicitly biased toward low-rank solutions, while Adam, starting from the same small initialization, is not.
・We trace the difference to the gauge symmetry of the loss, its invariance under $(U, V) \mapsto (UQ, VQ)$.
・Gradient flow's low-rank mechanism is available to an optimizer only if that optimizer is gauge-equiv
cs.LG updates on arXiv.org
The Luna Bound Propagator for Formal Analysis of Neural Networks
・arXiv:2603.23878v3 Announce Type: replace Abstract: The parameterized CROWN analysis, a.k.a., alpha-CROWN has emerged as a practically successful abstract interpretation method for neural network verification.
・However, existing implementations of alpha-CROWN are limited to Python, which complicates integration into existing DNN verifiers and long-term production-level systems.
・We introduce Luna, a new abstract-interp
The Verge
The messy politics behind Google’s big AI shakeup
・Google CEO Sundar Pichai at Google I/O.
・| Photo by Benjamin Fanjoy/Getty Images In the AI industry, Google prides itself on seeming like the adult in the room: quiet, stable, time-tested.
・On Wednesday, even as the company announced its largest AI org shakeup yet, Google and its leaders presented a unified front, keeping their messaging focused on how the changes tee up future success.
cs.LG updates on arXiv.org
The Neural Echo: A Signal Processing Perspective for Understanding Neural Networks
・arXiv:2608.04864v1 Announce Type: cross Abstract: We introduce the neural echo as a tool for understanding the behavior of neural networks.
・It generalizes the model-based concepts of impulse responses, diffusion echoes, and filter echoes to learning-based methods.
・It provides local, space-adaptive impulse responses and filter kernels for a neural network, its so-called echoes.
Hugging Face Papers
The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads
The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads
cs.LG updates on arXiv.org
The Price of Isolation: Estimating the Ecosystem Cost of Symmetric Two-Sided A/B Testing
・arXiv:2608.04432v1 Announce Type: cross Abstract: On two-sided content platforms, symmetric two-sided isolation (assigning matched fractions of creators and viewers to isolated treatment and control submarkets) is widely used for creator-side and cold-start experiments because it removes cross-arm marketplace interference.
・Isolation, however, thins each viewer's candidate catalog, and intuition suggests the resulting
cs.LG updates on arXiv.org
The RAIL Principles for Neurosymbolic AI: Reasoning, Assurances, Interfacing and Learning
・arXiv:2608.04285v1 Announce Type: cross Abstract: Neurosymbolic AI systems that integrate machine learning and symbolic reasoning are rapidly gaining attention.
・They complement the data-intensive statistical approaches of neural networks and language models with symbolic reasoning algorithms to function in high-stakes domains or in low-data regimes that characterize many real-world applications.
・We argue that the neu
cs.LG updates on arXiv.org
The Sample Complexity of Distributionally Robust PAC Learning under Cressie--Read Divergences
・arXiv:2608.04686v1 Announce Type: new Abstract: We study distributionally robust PAC learning for the $0$--$1$-loss, where adversarial perturbations of the data distribution are constrained by a Cressie--Read divergence of order $k>1$ and radius $\rho\geq 0$.
・For hypothesis classes with VC dimension $d$, we establish realizable and agnostic sample-complexity bounds tight up to constant and logarithmic factors, respec
cs.LG updates on arXiv.org
The Yokai Learning Environment: Tracking Beliefs Over Space and Time
・arXiv:2508.12480v3 Announce Type: replace-cross Abstract: The ability to cooperate with unknown partners is a central challenge in cooperative AI and widely studied in the form of zero-shot coordination (ZSC), which evaluates an algorithm by measuring the performance of independently trained agents when paired.
・The Hanabi Learning Environment (HLE) has become the dominant benchmark for ZSC, but recent work has achiev
WIRED
There’s No Good Way to Talk About Celebrities and Eating Disorders
・Online speculation over the weight loss of celebrities like Ariana Grande—even out of genuine concern—can make things worse for people struggling with eating disorders.
cs.LG updates on arXiv.org
TIDE: A Physically Diverse 3D Turbulence Benchmark Dataset for Advancing Scientific Machine Learning
・arXiv:2608.04222v1 Announce Type: cross Abstract: Turbulence is a central testbed for machine learning on physical dynamics because its governing laws are known exactly.
・However, most existing studies remain in 2D, while 3D turbulence has fundamentally different physics and is far more costly to simulate.
・Existing 3D resources also typically provide only one realization per configuration, making it difficult to disti
The Verge
TikTok blames ‘moderator error’ on slow response to Perez Hilton livestream
・TikTok says a "moderator error" delayed the removal of a livestream that appeared to show blogger Perez Hilton harming himself, as reported earlier by Wired.
・Jamie Favazza, a spokesperson for TikTok US, tells The Verge that the platform notified law enforcement about the situation and banned Hilton's account for violating its community guidelines.
・Hilton, or Mario Armando Lavandeira Jr., is known for the celebrity go
#LLMタグ
TIME誌が人間とAIクローラーに別ページを配信：AI回答を狙う「隠れ広告」とWebメディアの生存戦略
・導入 インターネットにおける情報の流通構造が、今まさに根本的な変革期を迎えています。これまでWeb上のメディアビジネスは、「人間がブラウザを開き、検索エンジンを経由してサイトを訪れ、ページ上の広告（バナーやテキスト広告）を見る」というサイクルによって成り立っていました。
Hugging Face Papers
ToolArtist: Tool-Using Unified Multimodal Models for Agentic Image Generation
ToolArtist: Tool-Using Unified Multimodal Models for Agentic Image Generation
WIRED
Total Wireless Promo Codes & Deals: 50% Off Select Plans
・Find great deals for Total Wireless, like 50% off select plans, and enjoy big savings.
Hugging Face Papers
Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
cs.LG updates on arXiv.org
Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
・arXiv:2608.05139v1 Announce Type: cross Abstract: Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule.
・We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs.
・Existing benchmarks of
Hugging Face Papers
Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes
Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes
cs.LG updates on arXiv.org
Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes
・arXiv:2608.05000v1 Announce Type: cross Abstract: Vision offers a critical axis for advancing foundation models, driving a shift towards natively unified multimodal pretraining.
・Despite this momentum, the design space and the fundamental mechanisms of how modalities interact during unified training remain underexplored.
・We provide empirical clarity through a systematic exploration of multimodal pretraining.
cs.LG updates on arXiv.org
Towards Trustworthy Hypergraph Neural Networks under Label Noise
・arXiv:2608.04377v1 Announce Type: new Abstract: Hypergraph neural networks (HGNNs) have demonstrated remarkable capabilities in processing complex higher-order relationships.
・However, their performance is highly dependent on labeled data, making them vulnerable to label noise.
・Despite advances in learning with label noise (LLN) and graph learning with label noise (GLN), noisy-label learning on hypergraphs remains und
cs.LG updates on arXiv.org
Towards Understanding Gradient Flow Dynamics of Homogeneous Neural Networks Beyond the Origin
・arXiv:2502.15952v3 Announce Type: replace Abstract: Recent works exploring the training dynamics of homogeneous neural network weights under gradient flow with small initialization have established that in the early stages of training, the weights remain small and near the origin, but converge in direction.
・Building on this, the current paper studies the gradient flow dynamics of homogeneous neural networks with loca
cs.LG updates on arXiv.org
Training Crossroads for Recurrent Vision Transformers: Recurrence, Neural ODEs, and Deep Supervision
・arXiv:2608.04879v1 Announce Type: new Abstract: Vision Transformers (ViTs) achieve strong image-recognition performance, but their parameter count grows linearly with depth when each block is independently parameterized.
・Single-block recurrent ViTs (bViT) remove this growth by repeatedly applying one shared block.
・Rather than proposing a new architecture, we fix a bViT and provide a controlled empirical characterizat
cs.LG updates on arXiv.org
Training-Free Hashing-Based Attention via Binary Principal Components
・arXiv:2608.04405v1 Announce Type: new Abstract: Long-context large language models (LLMs) are increasingly deployed in real-world applications, yet self-attention remains a major efficiency bottleneck -- especially during decoding -- due to the necessity of repeatedly processing ever-growing key-value (KV) caches.
・Existing sparse attention reduce computation by attending to fewer KV pairs, but often suffer from subst
cs.LG updates on arXiv.org
Transferable Dual-Stream Representations for Mesoscale-Preserving Sea Surface Temperature Downscaling
・arXiv:2608.04230v1 Announce Type: new Abstract: Deep learning models for scientific spatio-temporal downscaling often minimize reconstruction error while failing to preserve physically meaningful multi-scale structure.
・For sea surface temperature prediction, this can yield outputs that are numerically plausible yet overly smooth, missing mesoscale variability critical to regional ocean dynamics.
・Existing methods ofte
cs.LG updates on arXiv.org
Trident : How to Break Deep Reinforcement Learning Cyber Defenses (Agentic)
・arXiv:2608.04317v1 Announce Type: cross Abstract: Autonomous cyber defense systems based on Deep Reinforcement Learning (DRL) have attracted significant research attention, yet remain evaluated almost exclusively against static, heuristic red agents, leaving their robustness against adaptive threats critically understudied.
・Meanwhile, recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have impro
Hugging Face Papers
TriGlue: a Biology-Inspired Generative Model for Generating Molecular Glue-Induced Ternary Complex
TriGlue: a Biology-Inspired Generative Model for Generating Molecular Glue-Induced Ternary Complex
cs.LG updates on arXiv.org
Tropical Algebraic Geometry for Neuronal Representations: An Arakelov-Green Measure Based Descriptor for Graph Learning
・arXiv:2608.04460v1 Announce Type: new Abstract: The quantitative analysis of 3D neuronal morphologies requires capturing both graph topology and spatial geometry.
・Current message-passing Graph Neural Networks (GNNs) are bounded by the 1-Weisfeiler-Lehman (1-WL) test, limiting their ability to capture cycles induced by spatial proximities.
・To address this, we propose a training-free geometric prior based on tropical a
cs.LG updates on arXiv.org
TS2TabPFN: Time Series Classification and Extrinsic Regression through Feature Extraction and a Tabular Foundation Model
・arXiv:2608.04174v1 Announce Type: new Abstract: Time series data are ubiquitous in practical applications, where classification (TSC) and extrinsic regression (TSER) have emerged as essential tasks for obtaining value from temporal sequences.
・While the literature has seen significant progress through feature-based and deep learning models, existing methods often focus either on the quality of feature extraction or on
WIRED
Two Fossil Fuel Companies Are Betting Big on Data Centers
・Chevron and Williams are big winners in the race to power artificial intelligence as they build out gas-fired power plants and pipelines.
cs.LG updates on arXiv.org
Uncertainty-aware Predict-Then-Optimize Framework for Equitable Post-Disaster Power Restoration
・arXiv:2508.04780v2 Announce Type: replace Abstract: The increasing frequency of extreme weather events, such as hurricanes, highlights the urgent need for efficient and equitable power system restoration.
・Many electricity providers make restoration decisions primarily based on the volume of power restoration requests from each region.
・However, our data-driven analysis reveals significant disparities in request submis
cs.LG updates on arXiv.org
Understanding Fault Tolerance of Adversarially Robust Pruned Models
・arXiv:2608.04173v1 Announce Type: new Abstract: Deep neural networks (DNNs) deployed on resource-constrained neuromorphic hardware face three concurrent challenges: the need for model compression through pruning, vulnerability to adversarial input perturbations, and susceptibility to hardware-induced weight faults such as stuck-at-zero errors.
・While each of these factors has been studied in isolation, their combined
cs.LG updates on arXiv.org
Unforgettable Generalization in Language Models
・arXiv:2409.02228v2 Announce Type: replace Abstract: When language models (LMs) are trained to forget (or "unlearn'') a skill, how precisely does their behavior change?
・We study the behavior of transformer LMs in which tasks have been forgotten via fine-tuning on randomized labels.
・Such LMs learn to generate near-random predictions for individual examples in the "training'' set used for forgetting.
cs.LG updates on arXiv.org
Unifying quantum measurement constructions via a relative-entropy minimum change principle
・arXiv:2608.04055v1 Announce Type: cross Abstract: The minimum change principle provides an information-theoretic characterization of the Bayes reversal channel in classical probability theory and has recently been proposed as a framework for extending Bayes' rule to quantum information theory.
・Using quantum relative entropy, we investigate a minimum change principle for the setting of quantum statistical inference.
Hugging Face Papers
UniWorld-View: Large-Baseline View Synthesis via Video Diffusion Models
UniWorld-View: Large-Baseline View Synthesis via Video Diffusion Models
cs.LG updates on arXiv.org
Unscented KalmanNet: a hybrid deep learning filter with calibrated posterior covariance for nonlinear state estimation
・arXiv:2608.04201v1 Announce Type: new Abstract: State estimation for nonlinear dynamical systems is commonly performed with the Unscented Kalman filter (UKF), which propagates the state moments through deterministic sigma points and reports a posterior covariance at every step.
・In practice, however, unknown and time-varying noise statistics and model mismatch degrade both estimation accuracy and covariance calibratio
cs.LG updates on arXiv.org
Variational Bounds for Perceptron Learning from Structured Data
・arXiv:2608.04882v1 Announce Type: new Abstract: We introduce a variational approach to a finite-temperature continuous-spin perceptron trained on a Gaussian mixture.
・The model allows for a broad class of concave utilities and log-concave separable prior measures on the spins.
・By combining the interpolation method with log-concavity and concentration estimates, we derive lower and upper minimax variational bounds for
cs.LG updates on arXiv.org
Visual Representation Matters: Exploiting Temporal Differences in Video-to-Audio Generation
・arXiv:2608.04902v1 Announce Type: cross Abstract: Video-to-audio (V2A) generation extends image-to-audio generation (I2A) by introducing consecutive frames that provide essential temporal cues for audio synthesis.
・However, existing conditional diffusion-based V2A methods typically enhance visual conditioning with additional audio-visual supervision, acoustic structure prediction, or reasoning from large multimodal mo
Google DeepMind News
WeatherNext: AI model achieves breakthrough in forecasting cyclones
WeatherNext: AI model achieves breakthrough in forecasting cyclones
cs.LG updates on arXiv.org
Wedge Sampling: Efficient Tensor Completion with Nearly-Linear Sample Complexity
・arXiv:2602.05869v3 Announce Type: replace-cross Abstract: We introduce Wedge Sampling, a new non-adaptive sampling scheme for low-rank tensor completion.
・We study recovery of an order-$k$ low-rank tensor of dimension $n\times\cdots\times n$ from structured observations of its entries.
・Unlike the standard uniform entry model (i.e., i.i.d.
cs.LG updates on arXiv.org
What We Observe as LLM Behavior Can Be a Side-effect of Inference Backend
・arXiv:2608.04714v1 Announce Type: cross Abstract: Benchmark scores are reported as properties of a model, yet the inference framework used to produce them, such as HuggingFace, vLLM, or Ollama, are considered non-influential and their names and versions are almost never disclosed.
・In this work we investigate how much this choice can influence the model output.
・In a fully-crossed study (three instruction-tuned models
cs.LG updates on arXiv.org
When Does Latent Communication Pay? A Causal Audit of Relayed KV Caches in Multi-Agent LLMs
・arXiv:2608.04893v1 Announce Type: cross Abstract: Multi-agent LLM systems relay key--value caches instead of text and credit their gains to exchanged ``latent thoughts''.
・That credit is a claim about \emph{which} example's cache is relayed, not merely that one is.
・We audit it causally in released systems.
Hugging Face Papers
When Memory Lies: An Empirical Study of Spatial Memory Staleness in VLM Agents
When Memory Lies: An Empirical Study of Spatial Memory Staleness in VLM Agents
cs.LG updates on arXiv.org
When Modalities Fail to Tango: Conformal Backdoor Detection in Multimodal Contrastive Learning
・arXiv:2608.04052v1 Announce Type: cross Abstract: Backdoor attacks in multimodal contrastive learning (MCL) have garnered growing attention in recent years, as many downstream tasks critically depend on pre-trained MCL models.
・Existing detection-based defenses predominantly rely on the CLIPScore metric, under the assumption that poisoned pairs exhibit lower semantic similarity between the image and the caption.
cs.LG updates on arXiv.org
When More Becomes Less: Position-Dependent Repetition Effects in Language Models
・arXiv:2608.04021v1 Announce Type: cross Abstract: Cloze-style probes that vary how often a target token appears implicitly assume that more copies of a target affect prediction the same way regardless of where the readout slot sits.
・We show this assumption fails.
・Our two-probe design holds a repeated-target prefix fixed and varies only the readout position: the adjacent probe places the slot immediately after the rep
cs.LG updates on arXiv.org
When Proxy Prediction Becomes Equation Reconstruction: Diagnostics and Residual Learning for Factor-Derived Proxy Supervision
・arXiv:2608.04393v1 Announce Type: new Abstract: Scientific machine learning often relies on proxy targets computed from known domain factors when direct observations are limited.
・When those same factors are used as model inputs, however, high predictive accuracy may reflect reconstruction of the proxy-generating equation rather than robustness to degraded factor information.
・We study this problem in RUSLE-derived soi
Hugging Face Papers
When Teachers Mislead: Spurious-Signal-Aware On-Policy Distillation
When Teachers Mislead: Spurious-Signal-Aware On-Policy Distillation
WIRED
Whoop Promo Codes: 20% Off This August 2026
・Whether you're looking for a Whoop free trial, student discount, or military savings, our guide to Whoop promo codes will help you maximize your membership benefits.
・Stay on top of your fitness goals for less.
cs.LG updates on arXiv.org
Why Ranking Anomaly Detection Algorithms Isn't as Reliable as You May Think
・arXiv:2608.04613v1 Announce Type: new Abstract: Anomaly detection is a safety-critical machine learning problem with applications ranging from fraud detection to network intrusion prevention and industrial monitoring.
・Despite the large number of proposed anomaly detection algorithms, many novel methods claim state-of-the-art performance.
・However, many authors do so under benchmark settings that are not aligned with o
OpenAI News
Working with the American Psychological Association on youth mental health and AI
・OpenAI and the APA are launching a three-year partnership to develop guidance, resources, and safeguards for responsible AI use supporting youth mental health.
Hugging Face Papers
WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models
WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models
cs.LG updates on arXiv.org
WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models
・arXiv:2608.04964v1 Announce Type: cross Abstract: Interactive video world models are essential for long-horizon planning and exploration, yet they suffer from compounding errors.
・Post-training methods such as reinforcement learning (RL) can improve these models, but they hit a verification bottleneck: for arbitrary action sequences, no ground-truth future state exists to measure long-term drift.
・Our key insight is th
ITmedia NEWS 最新記事一覧
アプリが遅い原因をAIがトレースログから分析してくれる「Windows Performance Analyzer MCP」 Microsoftがプレビュー公開
・米Microsoftが、Windowsアプリケーションが遅くなる原因の調査分析をAIに依頼できるツール「Windows Performance Analyzer MCP」（WPA MCP）のアーリープレビューを発表しました。
#LLMタグ
お手軽LLMはじめてみた。その９（グラボがもう一枚届いたが。）
・Slot2に新しいRX470を Slot4に古いRX470を挿して起動。
・localai:~$ lspci -nn | grep -iE "Radeon"0000:17:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc.
・[AMD/ATI] Ellesmere [Radeon RX 470/480/570/570X/580/580X/590] [1002:67df] (rev cf)0000:17:00.1 Audio device [0403]: Advanced Micro Devices, Inc.
Qiita - 人気の記事
どんな通貨でも小数点以下の桁数をきれいに表示する方法がめっちゃ簡単だった
・きっかけ サブスク管理のようなアプリで「1アカウントあたりの単価」を通貨コード付きで表示する、という地味な機能を作っていた時のことなんですが 本来画面に出したいのは 0.28 USD や 1,200 JPY のはずなのに0.28 が 0,28 になってたり、1000 が ...
Zennの「大規模言語モデル」のフィード
ブラックボックスじゃないAIを作ろうとしたら、LLMの「ブレ」の正体が想定と違っていた話
・この記事の結論 LLMに何かを判定させるとき、同じ質問を複数回投げて、 答えのブレ具合で信頼度を測るというやり方があります。よく使われる手法です。
・私の環境では、そのブレの**100%が「1回目だけがズレる」**という現象でした。
・つまり測っていたのは「その質問の難しさ」ではなく、 **「1回目と2回目以降の実行状態の違い」**だった可能性が高い。
ITmedia NEWS 最新記事一覧
熊本地震で地面がずれた場所、35kmの線状に 衛星データで「変位境界」を判読 国土地理院
・国土地理院は8月5日、地球観測衛星「だいち2号」「だいち4号」のレーダーデータを解析し、令和8年熊本地震に伴って地表のずれが生じたとみられる境界（変位境界）を約35kmにわたり判読したと発表した。空中写真からは約2mの「右横ずれ」も読み取れたという。
ITmedia NEWS 最新記事一覧
個人開発「家系ラーメンマニア」で利用者急増 対応追いつかず一部停止 「より信頼していただけるアプリに」
・「家系ラーメンマニア」が店舗情報サービスを一時停止した。Xで紹介投稿が拡散し、利用者が約100人から4000人以上に急増。個人運営では情報更新や対応が追いつかず、体制整備後に再開する予定。投稿機能と地図機能は引き続き利用できる。
#AIタグ
自己紹介｜AIに「生成」させずに曲を作ろうとしている会社代表です
・はじめまして、木下です。
・コンテンツプロデュース会社・URAWAZAの代表をやっています。普段の仕事は、エンタメやスポーツまわりの企画、SNS運用、イベントの仕掛けづくり。会社としてはクライアントワークが中心ですが、このnoteは会社の看板を少し下ろした、個人の実験場です。
ITmedia NEWS 最新記事一覧
写真加工アプリ「SNOW」にステマで措置命令 報酬付きでX投稿依頼、広告表示なし 消費者庁
・消費者庁は8月6日、写真加工アプリ「SNOW」を運営する韓国企業と日本法人の2社に対し、広告と明示しないX投稿を第三者に依頼していたとして、景品表示法に基づく措置命令を出した。
#LLMタグ
弟子奮闘記：フィジカルＡＩ —— 弟子、データの土を面白がる
・我が弟子・ゆうころが松尾研フィジカルAI講座の第6回「実世界データの収集」に挑んだ。
・今回の講義は、一言で言えば土の話である。AIという花の下には、汚れた生データを学習可能な形へ磨き上げる、膨大で地味な工程が埋まっている。華やかなモデルの話ではない。土の話だ。