Wei Zhu

By integrating social attributes into relation modeling, RE-Rag achieves superior accuracy in predicting social media popularity, outperforming existing methods.

LLMZero uncovers that adaptive training strategies can boost RL performance by up to 140% by dynamically adjusting regularization parameters in response to training dynamics.

LLM reasoning often goes off the rails early, but targeted interventions at these critical junctures can dramatically improve performance.