Presentation and Link to Paper on Nested Mixture of Experts

Here, Junhyeok describes our work on training nested mixture of expert mdoels for data-efficient model-based reinforcement learning. In particular, we focus on learning mixture of expert representations of hybrid dynamical systems that engage in intermittent contacts, such as legged robots. You can access the paper by using the URL below:

http://proceedings.mlr.press/v144/ahn21a/ahn21a.pdf