Research group

Agent Systems and Reinforcement Learning

Constrained Policy Optimization

16 April 2019

A need to limit agent's action often arise in reinforcement learning, i.e. from the point of safety, when we are dealing with human-AI interactions.

At the seminar, we will discuss a recent paper "Constrained Policy Optimization", that adapts trust region optimization for constrained MDP to guarantee constraints on every step of the training process.

Paper: https://arxiv.org/abs/1705.10528

Speaker: Ilya Kaysin.

Presentation language: Russian.

Date and Time: April 16th, 18:30-20:00.

Place: Times, room 204.

Videos from previous seminars are available at http://bit.ly/MLJBSeminars

Resources