【AI最新動向 2026年2月20日】論文5件・GitHub5件

2026年2月20日

📝 この記事のポイント

🚀 AI技術の最新動向 - 2026年2月20日世界中から収集したAI・機械学習の最新情報をお届けします 📑 目次💻 注目のGitHubプロジェクトvixhal-baraiya/microgpt-cQwertyMcQwertz/monkeys-with-typewritersWB2024/Essentia-to-Metadatabenoitc/erlang-pythongreynewell/evaldriven.org📌 関連記事もチェック📚 最新研究論文1. Sink-Aware Pruning for Diffusion Language Models著者: Aidar Myrzakhan, Tianyi Li, Bowei GuoDiffusion Language Models (DLMs) incur high inference cost due to iterative denoising, motivating efficient pruning. Existing pruning heuristics largely inherited from autoregressive (AR) LLMs, typically preserve attention sink tokens because AR sinks serve as stable global anchors. We show that thi...論文を読む →2. CLEF HIPE-2026: Evaluating Accurate and Efficient Person-Place Relation Extraction from Multilingual Historical Texts著者: Juri Opitz, Corina Raclé, Emanuela BorosHIPE-2026 is a CLEF evaluation lab dedicated to person-place relation extraction from noisy, multilingual historical texts. Building on the HIPE-2020 and HIPE-2022 campaigns, it extends the series toward semantic relation extraction by targeting the task of identifying person--place associations in ...論文を読む →3. MARS: Margin-Aware Reward-Modeling with Self-Refinement著者: Payel Bhattacharjee, Osvaldo Simeone, Ravi TandonReward modeling is a core component of modern alignment pipelines including RLHF and RLAIF, underpinning policy optimization methods including PPO and TRPO. However, training reliable reward models relies heavily on human-labeled preference data, which is costly and limited, motivating the use of da...論文を読む →💻 注目のGitHubプロジェクト1. vixhal-baraiya/microgpt-cThe most atomic way to train and inference a GPT in pure, dependency-free C⭐ 197 stars | 🔀 38 forksリポジトリを見る →2. QwertyMcQwertz/monkeys-with-typewritersThe complete AI platform on a $3 microcontroller. Sub-millisecond inference. Zero hallucinations.⭐ 44 stars | 🔀 4 forksリポジトリを見る →3. WB2024/Essentia-to-MetadataIntelligent audio analysis and automatic genre/mood tagging using Essentia ML models⭐ 21 stars | 🔀 2 forksリポジトリを見る →4. benoitc/erlang-pythonExecute Python from Erlang using dirty NIFs with GIL-aware execution, rate limiting, and free-threading support⭐ 18 stars | 🔀 2 forksリポジトリを見る →5. greynewell/evaldriven.orgShip evals before you ship features.⭐ 12 stars | 🔀 4 forksリポジトリを見る →。

🚀 AI技術の最新動向 – 2026年2月20日

世界中から収集したAI・機械学習の最新情報をお届けします

📑 目次

💻 注目のGitHubプロジェクト
1. vixhal-baraiya/microgpt-c
2. QwertyMcQwertz/monkeys-with-typewriters
3. WB2024/Essentia-to-Metadata
4. benoitc/erlang-python
5. greynewell/evaldriven.org
📌 関連記事もチェック

📚 最新研究論文

1. Sink-Aware Pruning for Diffusion Language Models

著者: Aidar Myrzakhan, Tianyi Li, Bowei Guo

Diffusion Language Models (DLMs) incur high inference cost due to iterative denoising, motivating efficient pruning. Existing pruning heuristics largely inherited from autoregressive (AR) LLMs, typically preserve attention sink tokens because AR sinks serve as stable global anchors. We show that thi…

論文を読む →

2. CLEF HIPE-2026: Evaluating Accurate and Efficient Person-Place Relation Extraction from Multilingual Historical Texts

著者: Juri Opitz, Corina Raclé, Emanuela Boros

HIPE-2026 is a CLEF evaluation lab dedicated to person-place relation extraction from noisy, multilingual historical texts. Building on the HIPE-2020 and HIPE-2022 campaigns, it extends the series toward semantic relation extraction by targeting the task of identifying person–place associations in …

論文を読む →

3. MARS: Margin-Aware Reward-Modeling with Self-Refinement

著者: Payel Bhattacharjee, Osvaldo Simeone, Ravi Tandon

Reward modeling is a core component of modern alignment pipelines including RLHF and RLAIF, underpinning policy optimization methods including PPO and TRPO. However, training reliable reward models relies heavily on human-labeled preference data, which is costly and limited, motivating the use of da…

論文を読む →