Local Search GFlowNets - Enhancing Sampling Quality for Drug Discovery

Explore a 51-minute conference talk on Local Search GFlowNets presented by Minsu Kim from Valence Labs. Dive into the innovative training algorithm designed to enhance GFlowNets' sampling quality through local search methods. Learn how this approach improves the effectiveness of GFlowNets in various discovery tasks, particularly in drug discovery. Understand the challenges of vast combinatorial spaces and how local search techniques, including backtracking and trajectory reconstruction, target high-reward regions more efficiently. Examine the algorithm's structure, experimental results, and participate in a Q&A session. Gain insights into the potential applications of this advanced AI technique in the field of drug discovery and beyond.

- Intro + Background
- LS-GFN Algorithm
- Experiments
- Q&A