NUS PhD Student Wins 2025 Best Paper Award at NeurIPS 2025 for Advancing Long-Horizon AI Agents 

NUS PhD Student Wins 2025 Best Paper Award at NeurIPS 2025 for Advancing Long-Horizon AI Agents 

We’re proud to share that NUS Computer Science PhD student Zhou Zijian has won the Best Paper Award at the NeurIPS 2025 Workshop on Multi-Turn Interactions in Large Language Models, selected from 230 submitted papers. The winning paper, “MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents”, was developed in collaboration with SMART (Singapore-MIT Alliance for Research and Technology) M3S programme and researchers from MIT.

MEM1 introduces a new reinforcement learning approach that allows AI agents to operate more efficiently across long, multi-turn tasks. Instead of storing every past interaction – a common limitation that leads to growing memory use and slower inference – MEM1 trains an agent to maintain a compact, dynamic internal state that keeps only the information that truly matters for ongoing reasoning. This results in significantly lower memory usage and faster performance across several benchmark tasks. 

The team has made the code and model open-source to support further research in long-horizon interactive AI.

Join us in congratulating Zhou Zijian and all collaborators on this outstanding achievement!