Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning
Join us for an insightful discussion of the groundbreaking Search-R1 framework, presented by Bowen Jin, a fourth-year Ph.D. student in Computer Science at the University of Illinois Urbana-Champaign.
About the Paper
Search-R1 introduces a novel reinforcement learning framework that enables Large Language Models (LLMs) to autonomously generate search queries and seamlessly interleave reasoning with real-time retrieval. While LLMs excel at language tasks, they often struggle with complex reasoning and accessing current information. This innovative approach optimises LLM performance through multi-turn search interactions and a simple outcome-based reward function.
The research demonstrates significant performance improvements across multiple models:
26% improvement for Qwen2.5-7B
21% improvement for Qwen2.5-3B
10% improvement for LLaMA3.2-3B
About the Speaker
Bowen Jin is a PhD candidate advised by Prof. Jiawei Han and supported by the Apple PhD Fellowship and the Yunni and Maxine Pao Memorial Fellowship. His research focuses on the intersection of large language models, multimodal learning, and information networks, with particular interest in how foundational models can integrate various data types to solve real-world problems. Bowen’s current research interests include LLM agents, reasoning, and reinforcement learning, with publications in prestigious conferences, including ICLR, ICML, NeurIPS, KDD, SIGIR, ACL, COLM, and EMNLP.
Don’t miss this opportunity to explore the latest developments in LLM capabilities and search integration. The code for Search-R1 is available on GitHub.