Chen-Yu Lee
@chl260
Research Scientist @Google
"Deep Researcher with Test-Time Diffusion" This paper treats report writing as an iterative retrieval‑augmented diffusion process that can be enhanced by component‑wise self‑evolution. This demonstrates SoTA on multi‑hop search‑and‑reasoning benchmarks.
Deep Research Agents with Test-Time Diffusion Google keeps pushing on diffusion. This time, they apply diffusion to deep research agents, specifically the report generation process. It achieves a 69.1% win rate vs. OpenAI Deep Research on long-form research. My notes:
The #AISTATS 2025 Test of Time Award goes to ... 🥁 ... Chen-Yu Lee, Saining Xie, Patrick Gallagher, Zhengyou Zhang, Zhuowen Tu, for "Deeply Supervised Nets"! Congratulations!
📌 Grab the models, define your goal, and let Heterogeneous Swarms find multi-LLM systems for you! 📄 Paper: arxiv.org/abs/2502.04510 Joint work with @ZifengWang315 @palashiitkgp @yikewang_ @WeijiaShi2 Huang Xia @hamidpalangi @LukeZettlemoyer @tsvetshop @chl260 @tomaspfister
Happy to share that RevThink has been accepted to #NAACL2025 main conference! 🎉We also release the code and data 👇🧵 RevThink shows that LLMs can also benefit from reverse thinking (like we often do) 👉13.53% gains on 12 datasets (including MATH, ARC, ANLI, etc) + sample…
🚨 Reverse Thinking Makes LLMs Stronger Reasoners We can often reason from a problem to a solution and also in reverse to enhance our overall reasoning. RevThink shows that LLMs can also benefit from reverse thinking 👉 13.53% gains + sample efficiency + strong generalization!…
🚨 Reverse Thinking Makes LLMs Stronger Reasoners We can often reason from a problem to a solution and also in reverse to enhance our overall reasoning. RevThink shows that LLMs can also benefit from reverse thinking 👉 13.53% gains + sample efficiency + strong generalization!…
Searching for the ultimate LLM Knowledge distillation! Want one that excels in both task-specific and task-agnostic settings. Could it outperform others on varying data sizes and model initializations? Our Speculative Knowledge distillation might be the answer🚀 @Google…
👀 How to find a better adapted model? ✨ Let the models find it for you! 👉🏻 Introducing Model Swarms, multiple LLM experts collaboratively search for new adapted models in the weight space and discover their new capabilities. 📄 Paper: arxiv.org/abs/2410.11163
Speculative RAG splits Retrieval Augmented Generation tasks into two separate steps of drafting followed by verification to provide high-quality answer candidates, and substantially improve the quality and speed of the final output generation. Learn more →…
Interested in tabular understanding with LLMs? Check out our Chain-of-Table at #ICLR2024 ! The poster will be at Hall B #73 10:45-12:45 am. Feel free to ask questions and chat 😃
Stop by the #ICLR2024 Google booth today at 9:30 AM to hear @zlwang_cs, Lesly Miculicich, and @chl260 describe how Chain-of-Table enables step-by-step reasoning from table inputs by simplifying complex tables into more informative and manageable segments.
Why LLMs lost in the middle❓ 💡LLMs exhibit U-shape positional attention bias that dominates their generation behavior (often using leading/ending contexts in the response) 🚀By modeling and removing such bias, we hugely improve LLMs RAG performances! 📜: arxiv.org/abs/2406.16008