Jesse Farebrother
Papers
1
Total Citations
3
H-Index
1
About
Jesse Farebrother is a leading researcher in deep reinforcement learning (RL), whose work focuses on advancing the scalability and reliability of value-based methods. His most notable contribution, the paper "Stop Regressing: Training Value Functions via Classification for Scalable Deep RL" (2024), challenges the conventional use of mean squared error regression for training value functions. Instead, Farebrother proposes a classification-based approach, demonstrating that reframing value prediction as a distributional classification problem significantly improves stability and performance in large-scale neural networks. This work, already garnering 3 citations in its first year, has the potential to reshape how value functions are trained in deep RL, addressing long-standing issues with regression-based objectives. Farebrother’s research sits at the intersection of reinforcement learning, neural network optimization, and scalable AI, offering practical solutions for training agents in complex environments. His innovative perspective on value function learning marks him as a rising voice in the RL community, with implications for both academic research and real-world applications like robotics and game playing.
Research Focus
Key Achievements
Top Papers
- 1