Cross-domain knowledge transfer in policy gradient methods
Ability to generalize is one of the most discussed topics in machine learning. In reinforcement learning, ability to generalize is represented in training of multi-tasking agent. An agent that can solve several distinct tasks and transfer its experience to novel tasks.
At the seminar, we will discuss one approach to reusing agents' experience and go through a paper that proposes a method for knowledge transfer between environments.
Speaker: Oleg Svidchenko.
Presentation language: Russian.
Date and Time: February 12th, 18:30-20:00.
Place: Times, room 204.
Videos from previous seminars are available at http://bit.ly/MLJBSeminars