Topic: cs.LG

← Back to home

1. Diffusion Alignment as Variational Expectation-Maximization

   Paper • Feb 13, 2026 • arXiv • Zijing Ou, Jacob Si, Junyi Zhu, Yingzhen Li

   Diffusion alignment aims to optimize diffusion models for downstream objectives. While existing methods based on RL achieve success, they often suffer from reward over-optimization and mode collaps...

2. MEG-XL: Data-Efficient Brain-to-Text via Long-Context Pre-Training

   Paper • Feb 12, 2026 • arXiv • Dulhan Jayalath, Oiwi Parker Jones

   Decoding natural language from non-invasive brain recordings like Magnetoencephalography (MEG) remains a significant challenge due to the low signal-to-noise ratio and the scarcity of paired brain-...

3. High-accuracy sampling for diffusion models and log-concave distributions

   Paper • Feb 1, 2026 • arXiv • Fan Chen, Sinho Chewi, Constantinos Daskalakis, Alexander Rakhlin

   We present algorithms for diffusion model sampling which obtain δ-error in polylog(1/δ) steps, given access to eO(δ)-accurate score estimates in L2. This is an exponential improvement over all prev...

4. Learning Glioblastoma Tumor Heterogeneity Using Brain Inspired Topological Neural Networks

   Paper • Feb 12, 2026 • arXiv • Ankita Paul, Wenyi Wang

   Glioblastoma (GBM) is characterized by high intra-tumoral heterogeneity, which poses a significant challenge for diagnosis and treatment planning. We propose a brain-inspired Topological Neural Net...

5. Minimax M2.5: Scaling RL for Industrial-Grade Agentic AI

   Paper • Feb 16, 2026 • arXiv • MiniMax Research Team

   Training agents for industrial-scale deployment requires extreme stability and data throughput. We present Minimax M2.5, a model trained using a novel asynchronous RL architecture designed to proce...

6. MASPO: Robust and Sample-Efficient LLM Reasoning via Unified Policy Optimization

   Paper • Feb 19, 2026 • arXiv • Xiaoliang Fu, Jiaye Lin, Yangyi Fang

   Policy optimization for Large Language Models often suffers from gradient instability and reward signal unreliability, particularly in mathematical and verifiable reasoning tasks. We introduce MASP...

7. Fast KV Compaction via Attention Matching

   Paper • Feb 18, 2026 • arXiv • Adam Zweiger, Xinghong Fu, Han Guo, MIT Team

   Large Language Models struggle with memory overhead during long-context inference due to the linear growth of the Key-Value (KV) cache. We propose Attention Matching (AM), a framework for high-qual...

8. KLong: Training LLM Agents for Extremely Long-horizon Tasks

   Paper • Feb 19, 2026 • arXiv • Yue Liu, Zhiyuan Hu, Flood Sung

   Current LLM agents frequently fail in tasks requiring hundreds of steps due to error accumulation and context overflow. We introduce KLong, an agentic framework that utilizes 'Trajectory-Splitting ...

9. Identifying Intervenable and Interpretable Features via Orthogonality Regularization

   Paper • Feb 17, 2026 • arXiv • Moritz Miller, Florent Draye, Bernhard Schölkopf

   This paper addresses the fundamental challenge of 'feature disentanglement' in modern deep learning. We propose an Orthogonality Regularization technique to identify features that are both interpre...

10. KEPO: Knowledge-Enhanced Preference Optimization for Reinforcement Learning with Reasoning

    Paper • Feb 4, 2026 • arXiv • Fan Yang, Rui Meng, Trudi Di Qi, Ali Ezzati, Yuxin Wen

    Direct Preference Optimization (DPO) and its variants have revolutionized LLM alignment, yet they struggle when the preferred choice requires deep, multi-step reasoning. We introduce KEPO, a framew...