ENHANCING DEEP LEARNING WITH SYMBOLIC DOMAIN KNOWLEDGE

Abstract

Deep neural networks have brought significant advances to various tasks in machine learning and artificial intelligence. However, despite their effectiveness and flexibility, deep models have two drawbacks: low data efficiency and a lack of robustness.

Firstly, deep neural networks often require large amounts of training data. On the other hand, symbolic domain knowledge is often available in addition to data. The first part of this thesis aims to improve data efficiency by incorporating symbolic domain knowledge. We propose logic graph embedding frameworks, Logic Embedding Network with Semantic Regularization (LENSR) and Temporal-Logic Embedded Automata Framework (T-LEAF), which take propositional logic and linear temporal logic as inputs, respectively.

Secondly, recent work has shown that deep neural networks are vulnerable to adversarial attacks. In the second part of the thesis, we propose to leverage prior knowledge to defend against adversarial attacks in RL settings using the Knowledge-based Policy Recycling (KPR) framework.