
The author proposes a method called Di-SkilL that leverages Mixture of Experts to learn diverse skills through automatic curriculum learning. By optimizing per-expert context distributions using energy-based models, the approach enables efficient training and performance on challenging tasks.


coremsg

learning-diverse-skills-with-curriculum-reinforcement-learning-and-mixture-of-experts


Learning Diverse Skills with Curriculum Reinforcement Learning and Mixture of Experts
