repositories

ShellElastic: Scalable Elastic Pretraining with Simultaneous Maximal Updates and Hyperparameter Transfer (Submitted to ICLR 2027)

OPCD: Weak Critics Make Strong Learners — On-Policy Critique Distillation for Scalable Oversight (ICML 2026 AI4Math Workshop)

Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs (EMNLP 2026)

Two Heads are Better Than One: Test-time Scaling of Multi-agent Collaborative Reasoning (EMNLP 2026)

Learning from Teaching Regularization: Generalizable Correlations Should be Easy to Imitate (NeurIPS 2024)

LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025)

DARE: Difficulty-Adaptive Reinforcement Learning with Co-Evolved Difficulty Estimation

Codes for Visual Prompting Upgrades Neural Network Sparsification: A Data-Model Perspective (AAAI 2025)