Weiwei Sun

Most AI agents are tested in a bubble. But real ML breakthroughs happen in communities. We introduce CoMind, an research agent that learns from community knowledge. 📊 CoMind outperforms ~70% human teams in a CVPR 2025 workshop competition. 🧵👇

This work is done with amazing @__tm__157, @JunhongShen1, @sunweiwei12, @risteski_a, Yiming Yang, and @atalwalkar. We are excited about applying LLM techniques to solve more challenging PDEs in the future! 📝Paper: arxiv.org/abs/2505.08783 ⭐Code: github.com/LithiumDA/Code…

Can LLM solve PDEs? 🤯 We present CodePDE, a framework that uses LLMs to automatically generate solvers for PDE and outperforms human implementation! 🚀 CodePDE demonstrates the power of inference-time algorithms and scaling for PDE solving. More in 🧵: #ML4PDE #AI4Science

Thrilled to introduce FrontierCO — a benchmark of real, challenging (some unsolved) combinatorial optimization problems. It’s built to push frontier AI beyond toy tasks and toward advancing the boundaries of human problem-solving! Paper: arxiv.org/abs/2505.16952

🌟Get rid of the evaluation on synthetic toy problems and advance human intelligence like #AlphaEvolve! 🚀 Introducing FrontierCO — our new Machine Learning for Combinatorial Optimization benchmark featuring high-quality NP-hard instances from real-world applications and…

We invite you to explore our latest work on RAG. We conceptualize RAG as a multi-agent collaboration task, aiming to align and unify the optimization objectives of its various modules with the ultimate goal of generating high-quality answers.

This new approach for RAG modeling and optimization, MMOA-RAG, treats RAG as a multi-agent collaboration task. It uses MARL to simultaneously optimize multiple modules, aligning their objectives with the final goal of generating high-quality responses.

Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning Optimizes multiple RAG components jointly through multi-agent RL to align their goals toward generating high-quality answers. 📝 arxiv.org/abs/2501.15228 👨🏽‍💻 github.com/chenyiqun/MMOA…

Improving RAG through Multi-Agent RL This work treats RAG as a multi-agent cooperative task to improve answer generation quality. It models RAG components like query rewriting, document selection, and answer generation as reinforcement learning agents working together toward…

I will be presenting our paper “MAIR: A Massive Benchmark for Evaluating Instructed Retrieval” at #EMNLP2024! Date: Tuesday, Nov 12 Time: 14:00-15:30 Session 03: Resources and Evaluation 1 Paper: arxiv.org/abs/2410.10127 See you there!

The ability of an IR model to tackle various types of queries is THE MOST important problem when I was working on real search engine prodcution. Glad to complete this work with Weiwei and other co-authors that concludes many new insights to this problem. Check out our paper: