10 papers
SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking
Zhiwei Li, Lei Zhu, Hao Gu +6
Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context units (tokens or blocks) for e…
FlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience
Zixun Huang, Kishan Panaganti, Haitao Mi +1
A reasoning model can improve from its own on-policy experience, but this inner loop is fragile: terminal verifiers provide reliable yet sparse supervision, while dense same-model…
Stabilizing Efficient Reasoning with Step-Level Advantage Selection
Han Wang, Xiaodong Yu, Jialian Wu +4
Large language models (LLMs) achieve strong reasoning performance by allocating substantial computation at inference time, often generating long and verbose reasoning traces. While…
Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data
Zhenwen Liang, Yujun Zhou, Sidi Lu +3
Reinforcement Learning (RL) enhances LLM reasoning, yet a paradox emerges as models scale: strong base models saturate standard benchmarks (e.g., MATH), yielding correct but homoge…
Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge Exploration
Qifan Zhang, Dongyang Ma, Tianqing Fang +5
Most agents today ``self-evolve'' by following rewards and rules defined by humans. However, this process remains fundamentally dependent on external supervision; without human gui…
The Pensieve Paradigm: Stateful Language Models Mastering Their Own Context
Xiaoyuan Liu, Tian Liang, Dongyang Ma +4
In the world of Harry Potter, when Dumbledore's mind is overburdened, he extracts memories into a Pensieve to be revisited later. In the world of AI, while we possess the Pensieve-…