Jan 29, 2025
Is it better to train under noisy conditions to learn to generalize?
New work by Serena Bono and Spandan Madan shed light on the mechanisms underlying generalization in reinforcement learning agents.
This work led to discovering the indoor training effect whereby agents can improve their generalization performance under certain circumstances when trained under noise.
Serena Bono, Spandan Madan, Ishaan Grover, Mao Yasueda, Cynthia Breazeal, Hanspeter Pfister, Gabriel Kreiman. The Indoor-Training Effect: unexpected gains from distribution shifts in the transition function. arXiv: 2401.15856v2. Work to be presented at AAAI 2025.
Read the paper here.
New training approach could help AI agents perform better in uncertain conditions. MIT News.
Understanding the indoor training effect for AI agents. Tech Explorist.