Xuanye Xia
Publications
From Scoring to Acting: Outcome-Verified Comparative Self-Distillation for LLM Agents
Recent work on LLM agents is shifting from external capability elicitation to capability internalization, enabling agents to retain useful skills without retrieval at inference time. On-policy self-distillation (OPSD) offers a promising direction, but many existing methods typically supervise students by scoring actions along student-generated trajectories. Such supervision has two limitations: teacher preferences are not validated by environment outcomes, and action-level scores underuse information from student rollouts, teacher rollouts, and their behavioral relationship. We therefore advocate outcome-verified teacher supervision and comparative learning over teacher-student trajectories. Based on this view, we propose Outcome-Verified Comparative Self-Distillation (OVCSD). OVCSD organizes failed student rollouts into a prefix tree, adaptively invokes a skill-conditioned teacher from student-reached states, and retains only outcome-verified successful continuations. It then applies localized comparative learning at the first state-aligned divergence and distills the post-divergence teacher suffix to transfer completion behavior. Experiments on ALFWorld and WebShop across three model scales show that OVCSD consistently outperforms skill-free RL and existing self-distillation baselines, achieving up to 29.7 and 5.4 absolute success-rate gains over the strongest baselines on ALFWorld and WebShop, respectively, while adding less than 3% privileged interaction during training.
ClinicalReTrial: A Self-Evolving AI Agent for Clinical Trial Protocol Optimization
Clinical trial failure remains a central bottleneck in drug development, where minor protocol design flaws can irreversibly compromise outcomes despite promising therapeutics. Although cutting-edge AI methods achieve strong performance in predicting trial success, they are inherently reactive for merely diagnosing risk without offering actionable remedies once failure is anticipated. To fill this gap, this paper proposes ClinicalReTrial, a self-evolving AI agent framework that addresses this gap by casting clinical trial reasoning as an iterative protocol redesign problem. Our method integrates failure diagnosis, safety-aware modification, and candidate evaluation in a closed-loop, reward-driven optimization framework. Serving the outcome prediction model as a simulation environment, ClinicalReTrial enables low-cost evaluation of protocol modifications and provides dense reward signals for continuous self-improvement. To support efficient exploration, the framework maintains hierarchical memory that captures iteration-level feedback within trials and distills transferable redesign patterns across trials. Empirically, ClinicalReTrial improves 83.3% of trial protocols with a mean success probability gain of 5.7%, and retrospective case studies demonstrate strong alignment between the discovered redesign strategies and real-world clinical trial modifications.