University of Cambridge > > Machine Learning Reading Group @ CUED > Symmetries in Reinforcement Learning

Symmetries in Reinforcement Learning

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Elre Oldewage.

In recent years, there has been great interest in learning and exploiting symmetries of various machine learning problems. For example, this led to a generalization of CNNs originally designed for translation-invariant image classification in Euclidean space to much more general manifolds and graphs. The goal of this talk is to give an overview of approaches that specifically deal with symmetries in reinforcement learning problems, e.g. as commonly encountered in control, medicine or chemistry. In the first part of this talk, we will use group theory to characterize and formalize such symmetries using the notion of MDP homomorphisms. In the second part, we will present recent deep reinforcement learning methods for learning and exploiting MDP homomorphisms.

Recommended reading: 1) van der Pol, Elise, et al. “Plannable Approximations to MDP Homomorphisms: Equivariance under Actions.” arXiv preprint arXiv:2002.11963 (2020). 2) van der Pol, Elise, et al. “MDP homomorphic networks: Group symmetries in reinforcement learning.” Advances in Neural Information Processing Systems 33 (2020).

This talk is part of the Machine Learning Reading Group @ CUED series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity