Using Deep Neural Networks in Reinforcement Learning

Please enable javascript in your browser.

Sahaf Naeini, Alireza | 2017

923 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 50968 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Soleymani Baghshah, Mahdieh; Rabiei, Hamidreza
Abstract:
Reinforcement learning is a field of machine learning which is more similar to human training procedures.It uses reward signals to train an agent designed to act in that environment. Deep neural networks enhance the agent’s ability to determine and act better in its complex environment. Most previous works have addressed model-free agents, which ignore modeling details of the environment that in turn can be used to achieve better results. On the other hand, humans utilize a model-based approach in their decision-making process. They use their knowledge to predict the future and choose the action that leads them to a better state. To combine the benefits of model-based and model-free designs, we propose a compound network of reward and video frame prediction in order to estimate the model of the environment. We use this model to predict the future based on current state and desired action of the agent. We show that our approach can model the environment with less error than the existing model-based approaches in Atari environment,paving the way for future innovation in model-based agent research
Keywords:
Reinforcement Learning ; Deep Neural Networks ; Feature Extraction ; Video Prediction