Zesheng Shi

While recent self-training approaches have reduced reliance on human-labeled data for aligning LLMs, they still face critical limitations: (i) sensitivity to synthetic data quality, leading to instability and bias amplification in iterative training; (ii) ineffective optimization due to a diminis...

Reinforcement Learning with Verifiable Rewards (RLVR) is an emerging paradigm that significantly boosts a Large Language Model’s (LLM’s) reasoning abilities on complex logical tasks, such as mathematics and programming. However, we identify, for the first time, a latent vulnerability to backdoor...

Existing methods for enhancing the inductive reasoning of large language models (LLMs) at test-time typically depend on iterative self-refinement of hypotheses, which lacks explicit optimization guidance and effective error correction. This often results in superficial rewording and the accumulat...

While Large Language Models (LLMs) have demonstrated significant potential in Tool-Integrated Reasoning (TIR), existing training paradigms face significant limitations: Zero-RL suffers from inefficient exploration and mode degradation due to a lack of prior guidance, while SFT-then-RL is limited...

In this work, we introduce SkillWeave, a modular improvement framework that enables large language models to specialize under fixed memory budgets. SkillWeave partitions full capabilities of a general-purpose model into domain-specific skillpacks—lightweight, domain-specific delta modules—that re...