Thesis
Accelerate the learning speed of deep reinforcement learning by pre-training with non-expert human demonstrations
Washington State University
Master of Science (MS), Washington State University
2019
Handle:
https://hdl.handle.net/2376/102692
Abstract
Deep reinforcement learning (DRL) has achieved superior performance in complex sequential tasks by using deep neural networks as function approximators to learn directly from raw input images. However, learning directly from raw images is data inefficient. The agent must learn feature representation of complex states in addition to learning a policy. As a result, DRL typically suffers from slow learning speeds and often requires a prohibitively large amount of training time and data to reach reasonable performance, making it inapplicable to real-world settings where data is expensive. In this thesis, data efficiency in DRL can be improved by addressing one of the two learning goals, feature learning. Supervised learning is used to pre-train on a small set of non-expert human demonstrations and empirically evaluate this approach using the deep Q-network (DQN) and asynchronous advantage actor-critic algorithms (A3C) in the Atari domain. The results show significant improvements in learning speed, even when the provided demonstration is noisy and of low quality.
Metrics
26 File views/ downloads
68 Record Views
Details
- Title
- Accelerate the learning speed of deep reinforcement learning by pre-training with non-expert human demonstrations
- Creators
- Gabriel V. De la Cruz
- Contributors
- Shira Lynn Broschat (Degree Supervisor)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; [Pullman, Washington] :
- Identifiers
- 99900525094601842
- Language
- English
- Resource Type
- Thesis