<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0">
  <channel>
    <title>TECH Dashboard — AI Daily</title>
    <link>https://tech-dashboard.pages.dev</link>
    <description>AI コーディング/エコシステムの公式情報を毎日自動収集・要約</description>
    <language>ja</language>
    <lastBuildDate>Tue, 21 Apr 2026 08:37:57 GMT</lastBuildDate>
    <item>
      <title>Infrastructure Noise</title>
      <link>https://www.anthropic.com/engineering/infrastructure-noise</link>
      <guid isPermaLink="true">https://www.anthropic.com/engineering/infrastructure-noise</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:43 GMT</pubDate>
      <description>AnthropicのClaude Code評価において、推論・コーディング・エージェント的タスクでモデル性能を測定する際、インフラ起因のノイズ(ツールのタイムアウト、サンドボックス障害、レート制限等)が結果を歪める問題を分析。ノイズの特定・軽減手法を紹介し、信頼性の高いベンチマーク運用の重要性を論じる。</description>
      <category>claude</category>
      <category>anthropic</category><category>engineering</category><category>evaluation</category><category>benchmarking</category><category>claude-code</category><category>infrastructure</category>
    </item>
    <item>
      <title>Claude Opus 4 7</title>
      <link>https://www.anthropic.com/news/claude-opus-4-7</link>
      <guid isPermaLink="true">https://www.anthropic.com/news/claude-opus-4-7</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:42 GMT</pubDate>
      <description>Anthropicは最新モデル「Claude Opus 4.7」を発表した。前バージョンからコーディング、推論、エージェント的タスクの性能が向上し、フラッグシップモデルとして提供される。</description>
      <category>claude</category>
      <category>anthropic</category><category>claude</category><category>claude-opus</category><category>model-release</category>
    </item>
    <item>
      <title>A2UI v0.9: The New Standard for Portable, Framework-Agnostic Generative UI</title>
      <link>https://developers.googleblog.com/a2ui-v0-9-generative-ui/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/a2ui-v0-9-generative-ui/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleはA2UI v0.9を発表し、ポータブルでフレームワーク非依存の生成UI新標準を提示した。LLMが動的に生成するUIをあらゆるクライアントで統一的にレンダリングできる仕様で、エージェントとインターフェース間の相互運用性を高める。</description>
      <category>gemini</category>
      <category>google</category><category>a2ui</category><category>generative-ui</category><category>agents</category><category>standards</category>
    </item>
    <item>
      <title>MaxText Expands Post-Training Capabilities: Introducing SFT and RL on Single-Host TPUs</title>
      <link>https://developers.googleblog.com/maxtext-expands-post-training-capabilities-introducing-sft-and-rl-on-single-host-tpus/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/maxtext-expands-post-training-capabilities-introducing-sft-and-rl-on-single-host-tpus/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleはMaxTextにポストトレーニング機能を追加し、単一ホストTPU上で教師ありファインチューニング(SFT)と強化学習(RL)を実行可能にした。これによりLLMのカスタマイズを効率的なTPU環境で手軽に行えるようになる。</description>
      <category>gemini</category>
      <category>google</category><category>maxtext</category><category>tpu</category><category>sft</category><category>rlhf</category><category>fine-tuning</category>
    </item>
    <item>
      <title>New enhancements for merchant initiated transactions with the Google Pay API</title>
      <link>https://developers.googleblog.com/new-enhancements-for-merchant-initiated-transactions-with-the-google-pay-api/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/new-enhancements-for-merchant-initiated-transactions-with-the-google-pay-api/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Google Pay APIが加盟店主導取引(MIT)向けの機能強化を発表。サブスクリプションや定期課金などの継続決済シナリオに対応し、ネットワークトークンの活用で認証成功率向上やカード情報更新の自動化を実現する。</description>
      <category>gemini</category>
      <category>google</category><category>google-pay</category><category>payments</category><category>api</category>
    </item>
    <item>
      <title>Subagents have arrived in Gemini CLI</title>
      <link>https://developers.googleblog.com/subagents-have-arrived-in-gemini-cli/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/subagents-have-arrived-in-gemini-cli/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Gemini CLIにサブエージェント機能が追加された。専門特化した独立コンテキストを持つAIアシスタントを作成でき、特定タスクへの委任、並列実行、コンテキスト分離により複雑なワークフローを効率化できる。</description>
      <category>gemini</category>
      <category>google</category><category>gemini-cli</category><category>subagents</category><category>ai-agents</category>
    </item>
    <item>
      <title>Build Better AI Agents: 5 Developer Tips from the Agent Bake-Off</title>
      <link>https://developers.googleblog.com/build-better-ai-agents-5-developer-tips-from-the-agent-bake-off/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/build-better-ai-agents-5-developer-tips-from-the-agent-bake-off/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Google主催のAgent Bake-Offイベントで得られた、優れたAIエージェント構築のための5つの開発者向けヒントを紹介。Gemini活用やエージェント設計のベストプラクティスをまとめている。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>ai-agents</category><category>developer-tips</category>
    </item>
    <item>
      <title>Get ready for Google I/O: Livestream schedule revealed</title>
      <link>https://developers.googleblog.com/get-ready-for-google-io-livestream-schedule-revealed/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/get-ready-for-google-io-livestream-schedule-revealed/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Google I/Oのライブストリーム配信スケジュールが公開された。基調講演や開発者向けセッションの視聴予定を立てられるよう、各トピックの日程と時間帯が発表されている。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>google-io</category><category>livestream</category><category>event</category>
    </item>
    <item>
      <title>TorchTPU: Running PyTorch Natively on TPUs at Google Scale</title>
      <link>https://developers.googleblog.com/torchtpu-running-pytorch-natively-on-tpus-at-google-scale/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/torchtpu-running-pytorch-natively-on-tpus-at-google-scale/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleはPyTorchをTPU上でネイティブに動作させる新プロジェクト「TorchTPU」を発表した。PyTorch/XLAに代わり、Google規模でのTPU活用を可能にし、PyTorchユーザーがTPUの性能を直接引き出せるようにする。</description>
      <category>gemini</category>
      <category>google</category><category>pytorch</category><category>tpu</category><category>torchtpu</category><category>machine-learning</category>
    </item>
    <item>
      <title>Supporting Google Account username change in your app</title>
      <link>https://developers.googleblog.com/supporting-google-account-username-change-in-your-app/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/supporting-google-account-username-change-in-your-app/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Googleアカウントのユーザー名変更機能が導入されるため、開発者はアプリがメールアドレスではなく安定した識別子（subクレーム等）でユーザーを識別するよう対応が必要。Sign in with Google等での推奨実装方法を解説している。</description>
      <category>gemini</category>
      <category>google</category><category>google-account</category><category>authentication</category><category>sign-in-with-google</category>
    </item>
    <item>
      <title>Bring state-of-the-art agentic skills to the edge with Gemma 4</title>
      <link>https://developers.googleblog.com/bring-state-of-the-art-agentic-skills-to-the-edge-with-gemma-4/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/bring-state-of-the-art-agentic-skills-to-the-edge-with-gemma-4/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleはエッジデバイスでのエージェントAI向けに設計されたGemma 4を発表。最先端のエージェントスキルをオンデバイスで実現し、ローカル環境で高度なツール利用や自律的タスク処理を可能にする。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>open-model</category><category>release</category><category>gemma</category><category>edge-ai</category><category>agentic-ai</category><category>on-device</category>
    </item>
    <item>
      <title>Developer’s Guide to Building ADK Agents with Skills</title>
      <link>https://developers.googleblog.com/developers-guide-to-building-adk-agents-with-skills/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/developers-guide-to-building-adk-agents-with-skills/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Googleの開発者ブログが、Agent Development Kit (ADK)でSkillsを活用してエージェントを構築する方法を解説。Skillsは再利用可能な機能単位としてエージェントに組み込め、能力の拡張やモジュール化を容易にする。実装手順やベストプラクティスが紹介されている。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>tutorial</category><category>adk</category><category>agents</category><category>skills</category>
    </item>
    <item>
      <title>ADK Go 1.0 Arrives!</title>
      <link>https://developers.googleblog.com/adk-go-10-arrives/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/adk-go-10-arrives/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleがエージェントAI開発フレームワーク「ADK Go 1.0」を正式リリース。Goの性能と並行処理能力を活かし、本番環境向けのマルチエージェントシステム構築を可能にする。Python・Java版に続く3番目の安定版となる。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>release</category><category>adk</category><category>golang</category><category>ai-agents</category>
    </item>
    <item>
      <title>Boost Training Goodput: How Continuous Checkpointing Optimizes Reliability in Orbax and MaxText</title>
      <link>https://developers.googleblog.com/boost-training-goodput-how-continuous-checkpointing-optimizes-reliability-in-orbax-and-maxtext/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/boost-training-goodput-how-continuous-checkpointing-optimizes-reliability-in-orbax-and-maxtext/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>OrbaxとMaxTextにおける継続的チェックポイント機能を紹介。従来の永続的チェックポイント間隔でのみ保存する方式から、より頻繁にインメモリでチェックポイントを保存する方式に変更することで、大規模ML学習時の障害復旧時間を短縮し、学習のgoodput（有効稼働率）を向上させる。</description>
      <category>gemini</category>
      <category>google</category><category>orbax</category><category>maxtext</category><category>checkpointing</category><category>ml-training</category>
    </item>
    <item>
      <title>Announcing ADK for Java 1.0.0: Building the Future of AI Agents in Java</title>
      <link>https://developers.googleblog.com/announcing-adk-for-java-100-building-the-future-of-ai-agents-in-java/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/announcing-adk-for-java-100-building-the-future-of-ai-agents-in-java/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleはJava向けAgent Development Kit (ADK) 1.0.0の正式版をリリースした。JavaエコシステムでAIエージェントを構築するためのフレームワークで、マルチエージェント構成、ツール連携、Geminiなど各種LLM対応を提供し、本番環境での利用が可能となった。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>release</category><category>adk</category><category>java</category><category>ai-agents</category>
    </item>
    <item>
      <title>Closing the knowledge gap with agent skills</title>
      <link>https://developers.googleblog.com/closing-the-knowledge-gap-with-agent-skills/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/closing-the-knowledge-gap-with-agent-skills/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>Googleは、AIエージェントが特定タスクを実行するための知識やツールをパッケージ化する「エージェントスキル」の仕組みを紹介。モデルの再訓練なしに専門知識を付与でき、開発者はスキルを組み合わせて業務特化型エージェントを構築可能になる。</description>
      <category>gemini</category>
      <category>agent</category><category>google</category><category>agent-skills</category><category>ai-agents</category>
    </item>
    <item>
      <title>Jump to play: Building with Gemini &amp; MediaPipe</title>
      <link>https://developers.googleblog.com/jump-to-play-building-with-gemini-mediapipe/</link>
      <guid isPermaLink="true">https://developers.googleblog.com/jump-to-play-building-with-gemini-mediapipe/</guid>
      <pubDate>Tue, 21 Apr 2026 08:36:39 GMT</pubDate>
      <description>GoogleがGeminiとMediaPipeを組み合わせて開発したインタラクティブな体験「Jump to play」を紹介。MediaPipeによるリアルタイムの姿勢検出とGeminiのマルチモーダル理解を活用し、ユーザーの動きに応じて反応するブラウザベースのプレイ体験を構築する手法を解説している。</description>
      <category>gemini</category>
      <category>google</category><category>mediapipe</category><category>pose-detection</category><category>interactive-demo</category>
    </item>
    <item>
      <title>nightly: terminal_view: Don't try home_dir when working locally (#53071)</title>
      <link>https://github.com/zed-industries/zed/releases/tag/nightly</link>
      <guid isPermaLink="true">https://github.com/zed-industries/zed/releases/tag/nightly</guid>
      <pubDate>Tue, 21 Apr 2026 07:30:07 GMT</pubDate>
      <description>Zedのnightlyビルドにおいて、サイドバーのヘッダーにある省略記号メニュー内のワークスペース表示が調整されました。PR #54360による小規模なUI改善です。</description>
      <category>vscode</category>
      <category>editor</category><category>release</category><category>zed</category><category>sidebar</category><category>ui</category>
    </item>
    <item>
      <title>PlayStation’s age-gating restrictions are coming to UK consoles</title>
      <link>https://www.theverge.com/news/915448/sony-playstation-age-verification-uk-messaging-voice-chat</link>
      <guid isPermaLink="true">https://www.theverge.com/news/915448/sony-playstation-age-verification-uk-messaging-voice-chat</guid>
      <pubDate>Tue, 21 Apr 2026 07:11:34 GMT</pubDate>
      <description>ソニーはPlayStationの年齢確認機能を英国でも導入する。未成年ユーザーはメッセージ機能やボイスチャットなどの利用が制限され、英国のオンライン安全法に対応する措置となる。</description>
      <category>tech-news</category>
      <category>news</category><category>verge</category><category>playstation</category><category>sony</category><category>age-verification</category><category>uk</category><category>online-safety-act</category>
    </item>
    <item>
      <title>Google brings Pomelli in English to small businesses in Europe.</title>
      <link>https://blog.google/innovation-and-ai/models-and-research/google-labs/pomelli-in-europe/</link>
      <guid isPermaLink="true">https://blog.google/innovation-and-ai/models-and-research/google-labs/pomelli-in-europe/</guid>
      <pubDate>Tue, 21 Apr 2026 07:00:00 GMT</pubDate>
      <description>GoogleはAI活用マーケティングツール「Pomelli」の英語版を欧州の中小企業向けに提供開始した。企業のウェブサイトから独自のブランドアイデンティティを構築し、キャンペーン案やマルチチャネル向けコンテンツを自動生成できる。</description>
      <category>tech-news</category>
      <category>google</category><category>news</category><category>pomelli</category><category>marketing-ai</category><category>smb</category><category>europe</category>
    </item>
    <item>
      <title>v0.233.4-pre</title>
      <link>https://github.com/zed-industries/zed/releases/tag/v0.233.4-pre</link>
      <guid isPermaLink="true">https://github.com/zed-industries/zed/releases/tag/v0.233.4-pre</guid>
      <pubDate>Tue, 21 Apr 2026 04:40:00 GMT</pubDate>
      <description>Zedエディタのプレリリース版v0.233.4-preが公開されました。開発者向けの先行ビルドで、次期安定版に向けた修正や改善が含まれる通常の更新です。</description>
      <category>vscode</category>
      <category>agent</category><category>editor</category><category>release</category><category>zed</category><category>pre-release</category>
    </item>
    <item>
      <title>Dyson’s back with a travel-size Supersonic hairdryer</title>
      <link>https://www.theverge.com/gadgets/915165/dyson-supersonic-travel-hairdryer-gadgets-price</link>
      <guid isPermaLink="true">https://www.theverge.com/gadgets/915165/dyson-supersonic-travel-hairdryer-gadgets-price</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>Ten years ago, Dyson kicked off the hair gadget arms race with its $400 Supersonic hairdryer. Today, it's back with a slightly smaller and cheaper travel-size version. As the name suggests, the $299.9</description>
      <category>tech-news</category>
      <category>news</category><category>verge</category>
    </item>
    <item>
      <title>Multimodal Claim Extraction for Fact-Checking</title>
      <link>https://arxiv.org/abs/2604.16311</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16311</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文はファクトチェックのためのマルチモーダル主張抽出手法を提案。テキストと画像を統合し、検証すべき主張を自動抽出することで、従来のテキストのみに依存する手法の限界を克服する。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>fact-checking</category><category>multimodal</category><category>claim-extraction</category>
    </item>
    <item>
      <title>Cross-Family Speculative Decoding for Polish Language Models on Apple~Silicon: An Empirical Evaluation of Bielik~11B with UAG-Extended MLX-LM</title>
      <link>https://arxiv.org/abs/2604.16368</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16368</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>Apple Silicon上でポーランド語LLM「Bielik 11B」を対象に、異なるモデルファミリー間での投機的デコーディングを評価した論文。UAG(Universal Assisted Generation)を拡張したMLX-LMを用い、クロスファミリーのドラフトモデル活用による推論高速化を実証的に検証している。</description>
      <category>research</category>
      <category>arxiv</category><category>benchmark</category><category>paper</category><category>speculative-decoding</category><category>apple-silicon</category><category>mlx-lm</category><category>polish-nlp</category><category>bielik</category>
    </item>
    <item>
      <title>Brain-CLIPLM: Decoding Compressed Semantic Representations in EEG for Language Reconstruction</title>
      <link>https://arxiv.org/abs/2604.16370</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16370</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>Brain-CLIPLMは、EEG信号から言語を再構成する新手法を提案する。CLIPベースの意味圧縮表現を活用し、脳波から直接テキストを生成可能とする言語デコーディングの精度向上を目指す研究である。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>eeg</category><category>brain-decoding</category><category>clip</category><category>bci</category>
    </item>
    <item>
      <title>CFMS: Towards Explainable and Fine-Grained Chinese Multimodal Sarcasm Detection Benchmark</title>
      <link>https://arxiv.org/abs/2604.16372</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16372</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>中国語のマルチモーダル皮肉検出向けの説明可能かつ細粒度ベンチマークCFMSを提案する論文。画像とテキストを用いた皮肉検出タスクにおいて、検出結果だけでなく根拠となる説明や細かな分類を評価できるデータセットを構築した。</description>
      <category>research</category>
      <category>arxiv</category><category>benchmark</category><category>paper</category><category>sarcasm-detection</category><category>multimodal</category><category>chinese-nlp</category>
    </item>
    <item>
      <title>Foundational Study on Authorship Attribution of Japanese Web Reviews for Actor Analysis</title>
      <link>https://arxiv.org/abs/2604.16376</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16376</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>日本語Webレビューの著者帰属に関する基礎研究。悪質な投稿者(アクター)分析を目的とし、文体特徴や機械学習手法を用いて匿名レビューの書き手を推定する手法を検討している。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>authorship-attribution</category><category>japanese-nlp</category><category>web-reviews</category>
    </item>
    <item>
      <title>GoCoMA: Hyperbolic Multimodal Representation Fusion for Large Language Model-Generated Code Attribution</title>
      <link>https://arxiv.org/abs/2604.16377</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16377</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>GoCoMAは、LLM生成コードの帰属問題に対し、双曲空間上でコードの構文・意味・実行情報などのマルチモーダル表現を融合する手法を提案する。階層構造を効率的に捉え、既存手法より高精度にどのLLMがコードを生成したかを特定できる。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>code-attribution</category><category>hyperbolic-embedding</category><category>multimodal-fusion</category>
    </item>
    <item>
      <title>Reciprocal Co-Training (RCT): Coupling Gradient-Based and Non-Differentiable Models via Reinforcement Learning</title>
      <link>https://arxiv.org/abs/2604.16378</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16378</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、勾配ベースモデルと微分不可能なモデルを強化学習を介して結合する相互協調学習(RCT)を提案する。両モデルが互いに教師信号を与え合うことで、異なる性質のモデル間での知識転移と協調的な性能向上を実現する枠組みを示す。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>reinforcement-learning</category><category>co-training</category><category>knowledge-transfer</category>
    </item>
    <item>
      <title>Data Mixing for Large Language Models Pretraining: A Survey and Outlook</title>
      <link>https://arxiv.org/abs/2604.16380</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16380</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>大規模言語モデルの事前学習における「データ混合」手法を包括的に調査したサーベイ論文。オフライン/オンライン手法の分類、評価指標、既存研究の比較を整理し、今後の研究方向性と課題を示す。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>llm</category><category>pretraining</category><category>data-mixing</category><category>survey</category>
    </item>
    <item>
      <title>LiFT: Does Instruction Fine-Tuning Improve In-Context Learning for Longitudinal Modelling by Large Language Models?</title>
      <link>https://arxiv.org/abs/2604.16382</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16382</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文LiFTは、大規模言語モデルにおける指示ファインチューニングが、縦断的(時系列)モデリングの文脈内学習能力を向上させるかを検証する。指示調整モデルとベースモデルを比較し、長期的なデータパターン把握への影響を評価した。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>instruction-tuning</category><category>in-context-learning</category><category>longitudinal-modelling</category><category>llm-evaluation</category>
    </item>
    <item>
      <title>QU-NLP at QIAS 2026: Multi-Stage QLoRA Fine-Tuning for Arabic Islamic Inheritance Reasoning</title>
      <link>https://arxiv.org/abs/2604.16396</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16396</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>QIAS 2026共有タスク向けに、アラビア語イスラム相続法推論のため多段階QLoRAファインチューニングを適用した研究。段階的な学習戦略により、複雑な法的推論タスクで高精度を達成した。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>qlora</category><category>arabic-nlp</category><category>fine-tuning</category>
    </item>
    <item>
      <title>Measuring Representation Robustness in Large Language Models for Geometry</title>
      <link>https://arxiv.org/abs/2604.16421</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16421</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は大規模言語モデルの幾何学問題に対する内部表現の頑健性を測定する手法を提案する。問題文の意味を保った言い換えを与えた際に中間層の埋め込みがどの程度安定するかを分析し、LLMの推論の脆弱性を定量化する。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>llm</category><category>geometry</category><category>representation-learning</category><category>robustness</category>
    </item>
    <item>
      <title>Injecting Structured Biomedical Knowledge into Language Models: Continual Pretraining vs. GraphRAG</title>
      <link>https://arxiv.org/abs/2604.16422</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16422</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>生物医学知識を言語モデルに注入する二つの手法、継続事前学習とGraphRAGを比較した研究。構造化された医療知識グラフの活用法を検証し、それぞれの性能や適用場面の違いを評価している。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>biomedical-nlp</category><category>graphrag</category><category>continual-pretraining</category><category>knowledge-graph</category>
    </item>
    <item>
      <title>HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders</title>
      <link>https://arxiv.org/abs/2604.16430</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16430</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文HalluSAEは、スパースオートエンコーダ(SAE)を用いて大規模言語モデルの内部表現から幻覚に関連する特徴を抽出し、幻覚の検出を行う手法を提案する。既存手法より高精度に幻覚を識別でき、解釈可能性も向上させる。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>hallucination-detection</category><category>sparse-autoencoders</category><category>interpretability</category>
    </item>
    <item>
      <title>SynopticBench: Evaluating Vision-Language Models on Generating Weather Forecast Discussions of the Future</title>
      <link>https://arxiv.org/abs/2604.16451</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16451</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16451v1 Announce Type: new Abstract: Recent advances in visual-language models (VLMs) have led to significant improvements in a plethora of complex multimodal tasks like image captioning, r</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>EchoChain: A Full-Duplex Benchmark for State-Update Reasoning Under Interruptions</title>
      <link>https://arxiv.org/abs/2604.16456</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16456</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16456v1 Announce Type: new Abstract: Real-time voice assistants must revise task state when users interrupt mid-response, but existing spoken-dialog benchmarks largely evaluate turn-based i</description>
      <category>research</category>
      <category>arxiv</category><category>benchmark</category><category>paper</category>
    </item>
    <item>
      <title>Ethics of Care for Software Engineering</title>
      <link>https://arxiv.org/abs/2604.16303</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16303</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16303v1 Announce Type: new Abstract: Software engineering researchers repeatedly argue that the impact of their research on industrial practice, while desired and intended, is rarely achiev</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Results-Actionability Gap: Understanding How Practitioners Evaluate LLM Products in the Wild</title>
      <link>https://arxiv.org/abs/2604.16304</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16304</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16304v1 Announce Type: new Abstract: How do product teams evaluate LLM-powered products? As organizations integrate large language models (LLMs) into digital products, their unpredictable n</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Political and Ideological Pressure in Software Engineering Research: The Case of DEI Backlash</title>
      <link>https://arxiv.org/abs/2604.16305</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16305</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16305v1 Announce Type: new Abstract: Political and ideological pressures shape global research. Recently, these pressures have become particularly visible in research related to diversity, </description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Rethinking Artifact Evaluation for Software Engineering in the Age of Generative AI</title>
      <link>https://arxiv.org/abs/2604.16306</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16306</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16306v1 Announce Type: new Abstract: Peer review in software engineering research operates under tight time constraints, while generative AI has substantially reduced the human effort requi</description>
      <category>research</category>
      <category>arxiv</category><category>benchmark</category><category>paper</category>
    </item>
    <item>
      <title>AgentGuard: A Multi-Agent Framework for Robust Package Confusion Detection via Hybrid Search and Metadata-Content Fusion</title>
      <link>https://arxiv.org/abs/2604.16309</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16309</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16309v1 Announce Type: new Abstract: The proliferation of open-source software (OSS) has made software supply chains prime targets for attacks like Package Confusion, where adversaries publ</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Software Self-Extension with SelfEvolve: an Agentic Architecture for Runtime Code Generation</title>
      <link>https://arxiv.org/abs/2604.16314</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16314</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16314v1 Announce Type: new Abstract: Traditional self-adaptive systems automatically reconfigure existing components in response to changing requirements, but provide limited support for th</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Be a Partner, not a Bystander in Software Engineering Practice: Bridging the Gaps between Academia and Industry</title>
      <link>https://arxiv.org/abs/2604.16315</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16315</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16315v1 Announce Type: new Abstract: Software engineering conferences bring together thousands of academicians and software practitioners so that academic research and professional practice</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Software-Defined Vehicle Ecosystems in Transformation -- A Systematic Literature Review</title>
      <link>https://arxiv.org/abs/2604.16319</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16319</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16319v1 Announce Type: new Abstract: The automotive industry is shifting from hardware-centric development toward software-defined vehicles (SDVs), where software drives functionality, valu</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>How Robustly do LLMs Understand Execution Semantics?</title>
      <link>https://arxiv.org/abs/2604.16320</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16320</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16320v1 Announce Type: new Abstract: LLMs demonstrate remarkable reasoning capabilities, yet whether they utilize internal world models or rely on sophisticated pattern matching remains ope</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>LLM-Based Multi-Agent Systems for Code Generation: A Multi-Vocal Literature Review</title>
      <link>https://arxiv.org/abs/2604.16321</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16321</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16321v1 Announce Type: new Abstract: Large Language Models (LLMs) have enabled multi-agent systems to perform autonomous code generation for complex tasks. Despite the recent growth in rese</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Steerable Instruction Following Coding Data Synthesis with Actor-Parametric Schema Co-Evolution</title>
      <link>https://arxiv.org/abs/2604.16322</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16322</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16322v1 Announce Type: new Abstract: Interpreting and following human instructions is a critical capability of large language models (LLMs) in automatic programming. However, synthesizing l</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Beyond the 'Diff': Addressing Agentic Entropy in Agentic Software Development</title>
      <link>https://arxiv.org/abs/2604.16323</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16323</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16323v1 Announce Type: new Abstract: As autonomous coding agents become deeply embedded in software development workflows, their high operational velocity introduces a critical oversight ch</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>LLM4Log: A Systematic Review of Large Language Model-based Log Analysis</title>
      <link>https://arxiv.org/abs/2604.16359</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16359</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16359v1 Announce Type: new Abstract: Software systems generate massive, evolving, semi-structured logs that are central to reliability engineering and AIOps, yet difficult to analyze at sca</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Modelling GDPR-based Privacy Requirements with Software Engineering Diagrams: A Systematic Literature Review</title>
      <link>https://arxiv.org/abs/2604.16361</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16361</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16361v1 Announce Type: new Abstract: The application of the General Data Protection Regulation (GDPR) has significantly affected privacy requirements elicitation, modelling, and verificatio</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>A Systematic Review of MLOps Tools: Tool Adoption, Lifecycle Coverage, and Critical Insights</title>
      <link>https://arxiv.org/abs/2604.16371</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16371</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16371v1 Announce Type: new Abstract: Machine Learning Operations (MLOps) has become increasingly critical as more organisations move ML models into production. However, the growing landscap</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>DeepER-Med: Advancing Deep Evidence-Based Research in Medicine Through Agentic AI</title>
      <link>https://arxiv.org/abs/2604.15456</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15456</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>DeepER-Medは、医学分野でエビデンスに基づく研究を推進するためのエージェントAIフレームワーク。複数の専門エージェントが連携し、文献検索・評価・統合を自動化することで、臨床的な意思決定支援と体系的レビューの効率化を目指す。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>agentic-ai</category><category>medical-ai</category><category>evidence-based-medicine</category>
    </item>
    <item>
      <title>GIST: Multimodal Knowledge Extraction and Spatial Grounding via Intelligent Semantic Topology</title>
      <link>https://arxiv.org/abs/2604.15495</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15495</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>GISTは、インテリジェントな意味的トポロジーを用いて画像とテキストからマルチモーダルな知識抽出と空間的接地(グラウンディング)を同時に行う新しいフレームワークを提案する論文である。意味構造を活用することで精度向上を達成する。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>multimodal</category><category>grounding</category><category>semantic-topology</category>
    </item>
    <item>
      <title>Bureaucratic Silences: What the Canadian AI Register Reveals, Omits, and Obscures</title>
      <link>https://arxiv.org/abs/2604.15514</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15514</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文はカナダ政府のAI登録簿を分析し、公開されている情報が何を明らかにし、何を省略し、何を曖昧にしているかを検証する。官僚的な透明性制度の限界と、説明責任のギャップを指摘している。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>ai-governance</category><category>transparency</category><category>canada</category><category>public-sector-ai</category>
    </item>
    <item>
      <title>LACE: Lattice Attention for Cross-thread Exploration</title>
      <link>https://arxiv.org/abs/2604.15529</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15529</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>LACE(Lattice Attention for Cross-thread Exploration)は、複数の推論スレッド間で情報を共有するための格子状アテンション機構を提案する研究。スレッド横断的な探索を可能にし、並列推論の効率と精度を向上させる。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>attention</category><category>parallel-reasoning</category><category>transformer</category>
    </item>
    <item>
      <title>Preregistered Belief Revision Contracts</title>
      <link>https://arxiv.org/abs/2604.15558</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15558</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、事前登録型の信念修正契約という枠組みを提案し、エージェントが新情報を受け取る前に信念更新の方針を明示的にコミットする手法を定式化する。これにより信念変化の透明性と検証可能性を高めることを狙う。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>belief-revision</category><category>formal-epistemology</category><category>preregistration</category>
    </item>
    <item>
      <title>Subliminal Transfer of Unsafe Behaviors in AI Agent Distillation</title>
      <link>https://arxiv.org/abs/2604.15559</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15559</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文はAIエージェントの蒸留過程において、安全でない行動が潜在的に教師モデルから生徒モデルへ転移する「サブリミナル転移」現象を検証した研究である。明示的に有害データを除外しても、微細な統計的痕跡を通じて不安全行動が引き継がれる可能性を示す。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>ai-safety</category><category>distillation</category><category>agents</category><category>alignment</category>
    </item>
    <item>
      <title>Bilevel Optimization of Agent Skills via Monte Carlo Tree Search</title>
      <link>https://arxiv.org/abs/2604.15709</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15709</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、モンテカルロ木探索(MCTS)を用いてエージェントのスキルを二階層最適化する手法を提案する。上位層でスキル構造を探索し、下位層でパラメータを調整することで、複雑タスクにおけるエージェント性能を向上させる。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>mcts</category><category>agent-skills</category><category>bilevel-optimization</category><category>reinforcement-learning</category>
    </item>
    <item>
      <title>The World Leaks the Future: Harness Evolution for Future Prediction Agents</title>
      <link>https://arxiv.org/abs/2604.15719</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15719</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、世界が未来の情報を漏洩させるという観点から、進化的アプローチを活用した未来予測エージェントを提案する。エージェントは環境からの手がかりを収集・進化させ、予測精度を向上させる手法を示している。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>future-prediction</category><category>evolutionary-algorithms</category><category>agents</category>
    </item>
    <item>
      <title>LLM Reasoning Is Latent, Not the Chain of Thought</title>
      <link>https://arxiv.org/abs/2604.15726</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15726</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、LLMの推論能力は出力される思考連鎖(Chain of Thought)そのものではなく、潜在表現の中に存在すると主張する。CoTテキストは内部の潜在的推論過程の表層的な痕跡に過ぎず、モデルの真の推論機構を理解するには潜在空間の分析が必要であると論じている。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>llm</category><category>reasoning</category><category>chain-of-thought</category><category>latent-representations</category>
    </item>
    <item>
      <title>Structured Abductive-Deductive-Inductive Reasoning for LLMs via Algebraic Invariants</title>
      <link>https://arxiv.org/abs/2604.15727</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15727</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、代数的不変量を活用してLLMにアブダクション・演繹・帰納の構造化推論を行わせる手法を提案する。仮説生成を不変量探索に帰着させ、演繹的検証と帰納的一般化を組み合わせることで、推論の一貫性と検証可能性を高めることを目指す。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>llm-reasoning</category><category>abductive-reasoning</category><category>algebraic-invariants</category>
    </item>
    <item>
      <title>KWBench: Measuring Unprompted Problem Recognition in Knowledge Work</title>
      <link>https://arxiv.org/abs/2604.15760</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15760</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>KWBenchは、知識労働においてLLMが明示的な指示なしに問題を自発的に認識できるかを測定する新しいベンチマーク。実世界のタスクに潜む課題をモデルが気付けるかを評価し、従来の指示追従型評価を補完する。</description>
      <category>research</category>
      <category>arxiv</category><category>benchmark</category><category>paper</category><category>llm-evaluation</category><category>knowledge-work</category>
    </item>
    <item>
      <title>Stein Variational Black-Box Combinatorial Optimization</title>
      <link>https://arxiv.org/abs/2604.15837</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15837</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文はブラックボックス組合せ最適化のためのStein変分推論に基づく新手法を提案する。勾配情報が得られない離散探索空間において、粒子群を用いた分布近似で効率的に最適解を探索し、従来手法を上回る性能を示す。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>combinatorial-optimization</category><category>stein-variational</category><category>black-box-optimization</category>
    </item>
    <item>
      <title>Discover and Prove: An Open-source Agentic Framework for Hard Mode Automated Theorem Proving in Lean 4</title>
      <link>https://arxiv.org/abs/2604.15839</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15839</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>Lean 4におけるハードモードの自動定理証明のためのオープンソースのエージェント型フレームワーク「Discover and Prove」を提案。発見と証明を組み合わせた手法で、難易度の高い定理証明タスクに取り組む。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>lean4</category><category>theorem-proving</category><category>agentic-framework</category><category>open-source</category>
    </item>
    <item>
      <title>Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents</title>
      <link>https://arxiv.org/abs/2604.15877</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15877</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>LLMエージェントの経験蓄積手法を「記憶・スキル・ルール」という圧縮度の異なるスペクトルとして統一的に捉える枠組みを提案する論文。各形式の抽象化レベルと適用場面を整理し、エージェント設計における経験活用の指針を示す。</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category><category>llm-agents</category><category>memory</category><category>skills</category>
    </item>
    <item>
      <title>Towards Rigorous Explainability by Feature Attribution</title>
      <link>https://arxiv.org/abs/2604.15898</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.15898</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>本論文は、特徴量帰属による説明可能性に厳密な数学的基盤を与える試みを提案する。従来のヒューリスティックな手法を超え、説明の正確性や一貫性を保証する形式的枠組みを構築し、信頼性のあるAI解釈を目指す。</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category><category>explainability</category><category>feature-attribution</category><category>xai</category>
    </item>
    <item>
      <title>BASIS: Balanced Activation Sketching with Invariant Scalars for &quot;Ghost Backpropagation&quot;</title>
      <link>https://arxiv.org/abs/2604.16324</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16324</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16324v1 Announce Type: new Abstract: The activation memory required for exact backpropagation scales linearly with network depth, context length, and feature dimensionality, forming an O(L </description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>UniMamba: A Unified Spatial-Temporal Modeling Framework with State-Space and Attention Integration</title>
      <link>https://arxiv.org/abs/2604.16325</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16325</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16325v1 Announce Type: new Abstract: Multivariate time series forecasting is fundamental to numerous domains such as energy, finance, and environmental monitoring, where complex temporal de</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>A Discordance-Aware Multimodal Framework with Multi-Agent Clinical Reasoning</title>
      <link>https://arxiv.org/abs/2604.16333</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16333</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16333v1 Announce Type: new Abstract: Knee osteoarthritis frequently exhibits discordance between structural damage observed in imaging and patient-reported symptoms such as pain. This misma</description>
      <category>research</category>
      <category>agent</category><category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Preventing overfitting in deep learning using differential privacy</title>
      <link>https://arxiv.org/abs/2604.16334</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16334</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16334v1 Announce Type: new Abstract: The use of Deep Neural Network based systems in the real world is growing. They have achieved state-of-the-art performance on many image, speech and tex</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>SetFlow: Generating Structured Sets of Representations for Multiple Instance Learning</title>
      <link>https://arxiv.org/abs/2604.16362</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16362</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16362v1 Announce Type: new Abstract: Data scarcity and weak supervision continue to limit the performance of machine learning models in many real-world applications, such as mammography, wh</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Matched-Learning-Rate Analysis of Attention Drift and Transfer Retention in Fine-Tuned CLIP</title>
      <link>https://arxiv.org/abs/2604.16410</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16410</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16410v1 Announce Type: new Abstract: CLIP adaptation can improve in-domain accuracy while degrading out-of-domain transfer, but comparisons between Full Fine-Tuning (Full FT) and LoRA are o</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>CGCMA: Conditionally-Gated Cross-Modal Attention for Event-Conditioned Asynchronous Fusion</title>
      <link>https://arxiv.org/abs/2604.16411</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16411</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16411v1 Announce Type: new Abstract: We study asynchronous alignment, a first-class multimodal learning setting in which a dense primary stream must be fused with sporadic external context </description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Shifting the Gradient: Understanding How Defensive Training Methods Protect Language Model Integrity</title>
      <link>https://arxiv.org/abs/2604.16423</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16423</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16423v1 Announce Type: new Abstract: Defensive training methods such as positive preventative steering (PPS) and inoculation prompting (IP) offer surprising results through seemingly simila</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Functional Similarity Metric for Neural Networks: Overcoming Parametric Ambiguity via Activation Region Analysis</title>
      <link>https://arxiv.org/abs/2604.16426</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16426</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16426v1 Announce Type: new Abstract: As modern deep learning architectures grow in complexity, representational ambiguity emerges as a critical barrier to their interpretability and reliabl</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Non-Stationarity in the Embedding Space of Time Series Foundation Models</title>
      <link>https://arxiv.org/abs/2604.16428</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16428</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16428v1 Announce Type: new Abstract: Time series foundation models (TSFMs) are widely used as generic feature extractors, yet the notion of non-stationarity in their embedding spaces remain</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>(Sparse) Attention to the Details: Preserving Spectral Fidelity in ML-based Weather Forecasting Models</title>
      <link>https://arxiv.org/abs/2604.16429</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16429</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16429v1 Announce Type: new Abstract: We introduce Mosaic, a probabilistic weather forecasting model that addresses two principal sources of spectral degradation in ML-based weather predicti</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Dimensional Criticality at Grokking Across MLPs and Transformers</title>
      <link>https://arxiv.org/abs/2604.16431</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16431</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16431v1 Announce Type: new Abstract: Abrupt transitions between distinct dynamical regimes are a hallmark of complex systems. Grokking in deep neural networks provides a striking example --</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Sampling for Quality: Training-Free Reward-Guided LLM Decoding via Sequential Monte Carlo</title>
      <link>https://arxiv.org/abs/2604.16453</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16453</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16453v1 Announce Type: new Abstract: We introduce a principled probabilistic framework for reward-guided decoding in large language models, addressing the limitations of standard decoding m</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Multi-Label Phase Diagram Prediction in Complex Alloys via Physics-Informed Graph Attention Networks</title>
      <link>https://arxiv.org/abs/2604.16468</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16468</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16468v1 Announce Type: new Abstract: Accurate phase equilibria are foundational to alloy design because they encode the underlying thermodynamics governing stability, transformations, and p</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Positive-Only Drifting Policy Optimization</title>
      <link>https://arxiv.org/abs/2604.16519</link>
      <guid isPermaLink="true">https://arxiv.org/abs/2604.16519</guid>
      <pubDate>Tue, 21 Apr 2026 04:00:00 GMT</pubDate>
      <description>arXiv:2604.16519v1 Announce Type: new Abstract: In the field of online reinforcement learning (RL), traditional Gaussian policies and flow-based methods are often constrained by their unimodal express</description>
      <category>research</category>
      <category>arxiv</category><category>paper</category>
    </item>
    <item>
      <title>Pentagon pulls the plug on one of the military's most troubled space programs</title>
      <link>https://arstechnica.com/space/2026/04/pentagon-pulls-the-plug-on-one-of-the-militarys-most-troubled-space-programs/</link>
      <guid isPermaLink="true">https://arstechnica.com/space/2026/04/pentagon-pulls-the-plug-on-one-of-the-militarys-most-troubled-space-programs/</guid>
      <pubDate>Tue, 21 Apr 2026 02:27:52 GMT</pubDate>
      <description>米国防総省が、軍の最も問題を抱えた宇宙プログラムの一つを中止することを決定した。長年の開発遅延や予算超過が背景にあり、ペンタゴンは本計画から撤退する。</description>
      <category>tech-news</category>
      <category>ars-technica</category><category>news</category><category>pentagon</category><category>military-space</category><category>program-cancellation</category>
    </item>
    <item>
      <title>Who is John Ternus, the incoming Apple CEO?</title>
      <link>https://techcrunch.com/2026/04/20/who-is-john-ternus-the-incoming-apple-ceo/</link>
      <guid isPermaLink="true">https://techcrunch.com/2026/04/20/who-is-john-ternus-the-incoming-apple-ceo/</guid>
      <pubDate>Tue, 21 Apr 2026 01:02:04 GMT</pubDate>
      <description>Appleの次期CEOに就任予定と報じられているジョン・ターナス氏の人物像を紹介する記事。現在ハードウェアエンジニアリング担当シニアバイスプレジデントを務める同氏の経歴や役割、ティム・クック氏からの後継者としての位置付けに焦点を当てている。</description>
      <category>tech-news</category>
      <category>news</category><category>techcrunch</category><category>apple</category><category>john-ternus</category><category>ceo-succession</category><category>tim-cook</category>
    </item>
    <item>
      <title>v0.21.1-rc0</title>
      <link>https://github.com/ollama/ollama/releases/tag/v0.21.1-rc0</link>
      <guid isPermaLink="true">https://github.com/ollama/ollama/releases/tag/v0.21.1-rc0</guid>
      <pubDate>Tue, 21 Apr 2026 00:42:29 GMT</pubDate>
      <description>Ollamaのプレリリースバージョンv0.21.1-rc0が公開された。リリース候補版であり、詳細な変更内容は明記されていないが、次期安定版に向けたテスト用ビルドとして提供されている。</description>
      <category>local-llm</category>
      <category>ollama</category><category>release</category><category>release-candidate</category>
    </item>
    <item>
      <title>How to Ground a Korean AI Agent in Real Demographics with Synthetic Personas</title>
      <link>https://huggingface.co/blog/nvidia/build-korean-agents-with-nemotron-personas</link>
      <guid isPermaLink="true">https://huggingface.co/blog/nvidia/build-korean-agents-with-nemotron-personas</guid>
      <pubDate>Tue, 21 Apr 2026 00:40:10 GMT</pubDate>
      <description>NVIDIAがHuggingFaceブログで、合成ペルソナデータセット「Nemotron-Personas-Korea」を活用して韓国の実際の人口統計に基づくAIエージェントを構築する方法を紹介。地域・年齢・職業分布を反映したペルソナで、韓国向けLLMアプリの現実的な評価やデータ生成を可能にする。</description>
      <category>local-llm</category>
      <category>agent</category><category>huggingface</category><category>open-model</category><category>tutorial</category><category>nemotron</category><category>korean-nlp</category><category>synthetic-data</category><category>nvidia</category>
    </item>
    <item>
      <title>AIがチームの一員になった──Datadog Japan CS勉強会 #7 イベントレポート</title>
      <link>https://zenn.dev/datadog/articles/b33c5e2d8539c7</link>
      <guid isPermaLink="true">https://zenn.dev/datadog/articles/b33c5e2d8539c7</guid>
      <pubDate>Tue, 21 Apr 2026 00:15:05 GMT</pubDate>
      <description>Datadog Japan主催のCS勉強会#7のイベントレポート。テーマは「AIがチームの一員になった」で、AIをチームメンバーとして活用する事例や知見が共有された。カスタマーサクセス領域でのAI活用の現状と今後について議論された。</description>
      <category>research</category>
      <category>zenn</category><category>datadog</category><category>customer-success</category><category>event-report</category>
    </item>
    <item>
      <title>Tim Cook will still be Apple’s Trump whisperer</title>
      <link>https://www.theverge.com/policy/915422/tim-cook-apple-chairman-trump-policy</link>
      <guid isPermaLink="true">https://www.theverge.com/policy/915422/tim-cook-apple-chairman-trump-policy</guid>
      <pubDate>Tue, 21 Apr 2026 00:14:45 GMT</pubDate>
      <description>AppleのCEOティム・クックは、トランプ政権との関係維持を担う「トランプ対応役」として引き続き中心的な役割を果たす見通し。会長職などの役職変更の噂がある中でも、政治的折衝における彼の重要性が強調されている。</description>
      <category>tech-news</category>
      <category>news</category><category>verge</category><category>apple</category><category>tim-cook</category><category>trump</category><category>corporate-policy</category>
    </item>
    <item>
      <title>Claude CodeのUIをMCPで改善した話</title>
      <link>https://zenn.dev/ai_eris_log/articles/claude-code-design-mcp-20260421</link>
      <guid isPermaLink="true">https://zenn.dev/ai_eris_log/articles/claude-code-design-mcp-20260421</guid>
      <pubDate>Tue, 21 Apr 2026 00:09:58 GMT</pubDate>
      <description>Claude CodeのUIをMCP（Model Context Protocol）を活用して改善した取り組みを紹介する記事。MCPサーバーを導入することで、Claude Codeのユーザーインターフェースをより使いやすくカスタマイズした事例を解説している。</description>
      <category>research</category>
      <category>mcp-server</category><category>zenn</category><category>claude-code</category><category>mcp</category><category>ui</category>
    </item>
    <item>
      <title>Pairing geotechnical data with AI helps New Zealand build better</title>
      <link>https://news.microsoft.com/source/asia/features/pairing-geotechnical-data-with-ai-helps-new-zealand-to-build-better/</link>
      <guid isPermaLink="true">https://news.microsoft.com/source/asia/features/pairing-geotechnical-data-with-ai-helps-new-zealand-to-build-better/</guid>
      <pubDate>Mon, 20 Apr 2026 23:36:57 GMT</pubDate>
      <description>ニュージーランドでは地盤データとAIを組み合わせ、より安全で効率的な建設を実現する取り組みが進んでいる。Microsoftのクラウド・AI技術を活用し、地質調査データを統合・解析することで、インフラ計画や災害対策の精度向上が期待されている。</description>
      <category>tech-news</category>
      <category>microsoft</category><category>news</category><category>new-zealand</category><category>geotechnical</category><category>construction</category><category>azure-ai</category>
    </item>
    <item>
      <title>Anthropic takes $5B from Amazon and pledges $100B in cloud spending in return</title>
      <link>https://techcrunch.com/2026/04/20/anthropic-takes-5b-from-amazon-and-pledges-100b-in-cloud-spending-in-return/</link>
      <guid isPermaLink="true">https://techcrunch.com/2026/04/20/anthropic-takes-5b-from-amazon-and-pledges-100b-in-cloud-spending-in-return/</guid>
      <pubDate>Mon, 20 Apr 2026 23:10:27 GMT</pubDate>
      <description>AnthropicはAmazonから50億ドルの追加出資を受け、その見返りとしてAWSクラウドに1000億ドルを支出することを約束した。両社の提携関係がさらに深化し、計算資源の大規模確保につながる。</description>
      <category>tech-news</category>
      <category>news</category><category>techcrunch</category><category>anthropic</category><category>amazon</category><category>aws</category><category>investment</category>
    </item>
    <item>
      <title>vscode copilot備忘録用</title>
      <link>https://qiita.com/kchayama2008/items/a473271aa408adeb68dd</link>
      <guid isPermaLink="true">https://qiita.com/kchayama2008/items/a473271aa408adeb68dd</guid>
      <pubDate>Mon, 20 Apr 2026 23:09:13 GMT</pubDate>
      <description>VSCodeのGitHub Copilotに関する個人的な備忘録記事。使用上のメモや設定のTipsをまとめたもの。</description>
      <category>copilot</category>
      <category>qiita</category><category>vscode</category><category>memo</category>
    </item>
    <item>
      <title>施策の効果検証、まだ手作業でやってませんか？ — 施策×KPIをGraph RAG+MCPで探索可能にした話</title>
      <link>https://zenn.dev/aircloset/articles/7a0b06cb2a35d8</link>
      <guid isPermaLink="true">https://zenn.dev/aircloset/articles/7a0b06cb2a35d8</guid>
      <pubDate>Mon, 20 Apr 2026 23:02:05 GMT</pubDate>
      <description>エアクローゼットが施策とKPIの関係をGraph RAGとMCPで探索可能にした事例を紹介。従来手作業だった施策の効果検証を、ナレッジグラフ化しLLMから自然言語で問い合わせられる仕組みに刷新し、効率化を実現した。</description>
      <category>research</category>
      <category>mcp-server</category><category>rag</category><category>zenn</category><category>graph-rag</category><category>mcp</category><category>kpi</category><category>knowledge-graph</category>
    </item>
    <item>
      <title>SDD疲れの処方箋 ── AI時代の設計判断はADR 1枚で残す</title>
      <link>https://zenn.dev/kosk_t/articles/adr-lightweight-decision-records-for-ai</link>
      <guid isPermaLink="true">https://zenn.dev/kosk_t/articles/adr-lightweight-decision-records-for-ai</guid>
      <pubDate>Mon, 20 Apr 2026 23:00:11 GMT</pubDate>
      <description>AI時代の仕様駆動開発(SDD)で設計ドキュメントが肥大化し疲弊する問題に対し、設計判断をADR(Architecture Decision Record)1枚に軽量に記録する手法を提案する記事。ADRによる判断の可視化と保守性向上を論じている。</description>
      <category>research</category>
      <category>zenn</category><category>adr</category><category>spec-driven-development</category><category>software-design</category>
    </item>
    <item>
      <title>「今、隣のメンバーがどんなAI活用をしているのか？」AIエージェントとのやり取りをSlackに集約したら、AI活用が急加速した話</title>
      <link>https://zenn.dev/elements/articles/e13ffaa8876b69</link>
      <guid isPermaLink="true">https://zenn.dev/elements/articles/e13ffaa8876b69</guid>
      <pubDate>Mon, 20 Apr 2026 23:00:10 GMT</pubDate>
      <description>AIエージェントとのやり取りを個人チャットに閉じず、Slackチャンネルに集約することで、チームメンバーのAI活用事例を相互に可視化。他者のプロンプトや使い方を参考にできる環境を整えたところ、組織全体のAI活用が急速に進展したという事例紹介。</description>
      <category>research</category>
      <category>zenn</category><category>slack</category><category>ai-agent</category><category>team-collaboration</category>
    </item>
    <item>
      <title>CLAUDE.md の肥大化を 3 層構造で 83% 軽くした — 実測と試行錯誤の記録</title>
      <link>https://zenn.dev/pepabo/articles/claude-code-rules-skills-split</link>
      <guid isPermaLink="true">https://zenn.dev/pepabo/articles/claude-code-rules-skills-split</guid>
      <pubDate>Mon, 20 Apr 2026 22:50:40 GMT</pubDate>
      <description>CLAUDE.mdの肥大化問題に対し、ルール・スキル・コンテキストの3層構造に分割することで83%の軽量化を実現した事例。実測値と試行錯誤のプロセスを記録し、Claude Codeの運用効率化手法を紹介している。</description>
      <category>research</category>
      <category>zenn</category><category>claude-code</category><category>claude-md</category><category>prompt-engineering</category>
    </item>
    <item>
      <title>The Lenovo Legion Go S is RAMageddon’s latest victim</title>
      <link>https://www.theverge.com/games/915278/lenovo-legion-go-s-price-hike-discontinued-ramageddon</link>
      <guid isPermaLink="true">https://www.theverge.com/games/915278/lenovo-legion-go-s-price-hike-discontinued-ramageddon</guid>
      <pubDate>Mon, 20 Apr 2026 22:49:42 GMT</pubDate>
      <description>Lenovo Legion Go S がRAM価格高騰(RAMageddon)の影響で値上げされ、一部モデルは販売終了となった。メモリ不足と価格上昇が携帯ゲーミングPC市場に波及している。</description>
      <category>tech-news</category>
      <category>news</category><category>verge</category><category>lenovo</category><category>legion-go-s</category><category>handheld-pc</category><category>ram-shortage</category>
    </item>
    <item>
      <title>1 件のバグ報告が暴いた『並列の嘘』と、外に出した並列ラッパー──Clade v1.21.2 &amp; clade-parallel v0.1</title>
      <link>https://zenn.dev/satoh_y_0323/articles/a7d691dba8eb12</link>
      <guid isPermaLink="true">https://zenn.dev/satoh_y_0323/articles/a7d691dba8eb12</guid>
      <pubDate>Mon, 20 Apr 2026 22:45:17 GMT</pubDate>
      <description>Cladeのv1.21.2では、バグ報告で発覚した『並列実行のふり』を修正し、真の並列処理を実現。また並列処理機能を外部ライブラリclade-parallel v0.1として切り出し公開した。</description>
      <category>research</category>
      <category>zenn</category><category>clade</category><category>parallel-processing</category><category>bug-fix</category>
    </item>
    <item>
      <title>M365 Copilotを企業で安全に使うためのCCSとは何か、AI時代の企業内データ保護と統制</title>
      <link>https://zenn.dev/syoshida07/articles/2f70e9a756396e</link>
      <guid isPermaLink="true">https://zenn.dev/syoshida07/articles/2f70e9a756396e</guid>
      <pubDate>Mon, 20 Apr 2026 22:38:26 GMT</pubDate>
      <description>企業でMicrosoft 365 Copilotを安全に活用するためのCCS(Copilot Control System)について解説。AI時代におけるデータガバナンス、アクセス制御、機密情報保護、利用統制の仕組みを整理し、社内データ漏洩リスクを抑える運用方法を示している。</description>
      <category>research</category>
      <category>zenn</category><category>m365-copilot</category><category>data-governance</category><category>enterprise-security</category>
    </item>
    <item>
      <title>【AI音声詐欺 対策ガイド2026】AI 音声詐欺の技術的構造と、個人・家族が取れる対策の全体像</title>
      <link>https://zenn.dev/ai_japan_index/articles/850d914b962d4e</link>
      <guid isPermaLink="true">https://zenn.dev/ai_japan_index/articles/850d914b962d4e</guid>
      <pubDate>Mon, 20 Apr 2026 22:31:04 GMT</pubDate>
      <description>AI音声詐欺の技術的仕組みと、個人や家族が取るべき対策を2026年版ガイドとして解説する記事。音声合成技術の悪用手口を整理し、家族間の合言葉設定や本人確認手順など実践的な防御策の全体像を示している。</description>
      <category>research</category>
      <category>zenn</category><category>voice-cloning</category><category>scam-prevention</category><category>deepfake</category>
    </item>
  </channel>
</rss>
