Xin Guo

This work introduces ReDraft (Reference-Driven Revision and Fine-Tuning), which obtains both from the model's own failures: using an expert response only as a reference, it has the model revise its own incorrect rollout, keeps the revision only if a verifier accepts it, and fine-tunes on what survives.

CORAL automates the continual optimization of recommender systems, achieving improved user engagement and reduced costs without human intervention.

Low-probability tokens disproportionately influence model updates, and a simple reweighting strategy can significantly enhance performance without sacrificing generalization.

IACM-RL reduces infinite loops and stale context errors by proactively managing dynamic user intents, setting a new standard for robust tool invocation.

Deterministic policies can significantly enhance the stability and efficiency of reinforcement learning in complex mean field control problems, outperforming traditional stochastic approaches.

Even state-of-the-art language models struggle significantly in real-world tasks, exposing critical shortcomings in their deployment readiness.

Time-inconsistent control problems can be tackled effectively with a novel two-stage reinforcement learning algorithm that learns equilibrium policies, bridging theory and practical applications in finance.