Towards More Human Like Reinforcement Learning
Name
Zurabi Isakadze
Abstract
Making machines more intelligent can potentially make human life easier. A lot of research has gone into the field of artificial intelligence (AI) since the creation of first computers. However, today’s systems still lag behind humans’ general ability to think and learn. Reinforcement Learning (RL) is a framework where software agents learn by interaction with an environment. We investigate possibilities to use observations about human intelligence to make RL agents smarter. In particular, we tried several methods: 1) To use “Tagger” - an unsupervised deep learning framework for perceptual grouping, to learn more usable abstract relationships between objects; 2) Make one RL algorithm (A3C) more data efficient to learn faster; 3) To conduct these experiments, we built a web based RL dashboard based on visualization tool - visdom. Finally, we provide some concrete challenges to work on in the future.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Jaan Aru, Raul Vicente
Defence year
2017