Saravan Rajmohan

Recasting Transformer Layers as Energy Models

Pretrain Value, Not Reward: Decoupled Value Policy Optimization

DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems

Text2Grad: Reinforcement Learning from Natural Language Feedback

WarriorMath: Empowering Mathematical Reasoning for Large Language Models via Expert Battles

ACON: Optimizing Context Compression for Long-horizon LLM Agents

Memory-Augmented Personalized Retrieval for Long-Context Egocentric Video

GUI-360° : A Comprehensive Dataset and Benchmark for Computer-Using Agents

G-KV: Decoding-Time KV Cache Eviction with Global Attention

Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs

Duet: Joint Exploration of User–Item Profiles

VEM: Environment-Free Exploration for Training GUI Agent with Value Environment Model

Distill Not Only Data but Also Rewards: Can Smaller Language Models Surpass Larger Ones?

Synergistic Weak-Strong Collaboration by Aligning Preferences

Unveiling Context-Aware Criteria in Self-Assessing LLMs