Interpretable Reinforcement Learning with the Regression Tsetlin Machine

Varma, Varun Ravi (2021) Interpretable Reinforcement Learning with the Regression Tsetlin Machine. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text Master_Thesis_VRVarma (1).pdf Download (1MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (100kB)

Abstract

As Artificial Intelligence (AI) methods are being widely adopted by industries in multiple domains, there is an increasing concern in the ethics and transparency of decisions made by AI tools. Although these black box agents improve the speed of decision making, we have seen quite a few scenarios where the agent displays clear biases in the decisions made. Hence, it has become increasingly important that the AI agent behaviour is transparent. Such transparency can be brought about either by making the agent's decision making process interpretable or by utilizing external tools to view the decision making process (white box equations or visualization tools). Widely used RL agents utilize Deep Neural Networks (DNNs) to make decisions, and this makes the agent behaviour difficult to explain, since the complexity of DNNs increase with each layer. The Tsetlin Machine is an alternative learning mechanism to deep neural networks, which has been shown to have comparable accuracy to state of the art machine learning algorithms with the additional advantages of interpretability and computational simplicity. In this project, we aim to implement an interpretable reinforcement learning framework utilizing Tsetlin Machines and Q-learning. We will then test this agent out on two classical control environments and try to understand how patterns in the input lead to decisions and also compare its performance with existing reinforcement learning methods.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Mohades Kasaei, S.H.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	07 Feb 2022 10:30
Last Modified:	07 Feb 2022 10:30
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/26483

Actions (login required)

View Item