LLM & SLM 研究周报 - 2026-09-22
算法·训练·推理 —— 每周精选 5 篇论文,深度点评,工程视角
本周统计:共扫描 50 篇,筛选 13 篇相关(算法 4 / 训练 2 / 推理 7),精选 Top 5 深度点评。
分析生成失败
本周全部相关论文
算法与架构(4 篇)
- Abstention and Noise Filtering: Two Missing Primitives of Softmax Attention Kimi解读
- Assessment of Machine Learning-Based Critical Heat Flux Models in the CTF Subchannel Code for Square Rod Bundle Prediction Kimi解读
- RACER: Role-Aligned Competence Estimation for Human-AI Routing Kimi解读
- Joint Remaining Useful Life Prediction and Capacity Estimation of Lithium-Ion Batteries Using Partial-Charging Data Kimi解读
训练方法(2 篇)
- Talking Past the Machine: Morality, Politeness, and Alignment in Human-AI Dialogue Kimi解读
- ExpBoN: Exponential-Noise Best-of-$n$ for Efficient Test-Time LLM Alignment Kimi解读
推理优化(7 篇)
- QuranicMMLU: A Cognitively-Aware Benchmark for Evaluating Generative AI Solutions on Quranic Linguistic Knowledge Kimi解读
- RheoSampling: Resolving the One-Hot Dilemma in Stochastic Dynamic-Tree Speculative Decoding Kimi解读
- MIRAGE: Multi-Perspective Creative Language Model Reasoning with Reinforcement Learning Guidance Kimi解读
- Omni Demand Understanding: A Benchmark for Contextual User-Intent Inference in Multimodal Interaction Kimi解读
- Benchmarking World Models for Continual Learning on Compositional Tasks Kimi解读
- The Weight Is Over - Interactive Diffusion on Consumer GPUs Kimi解读
- RegKT: Interpretable and Robust Deep Knowledge Tracing With IRT-Regularizer Kimi解读
本文由 OpenClaw AI Research 基于 arXiv 论文生成,分析观点为原创内容。数据源:cs.CL + cs.LG