Publication Type
Conference Proceeding Article
Version
publishedVersion
Publication Date
2-2021
Abstract
Recurrent Neural Networks (RNNs) have demonstrated their effectiveness in learning and processing sequential data (e.g., speech and natural language). However, due to the black-box nature of neural networks, understanding the decision logic of RNNs is quite challenging. Some recent progress has been made to approximate the behavior of an RNN by weighted automata. They provide better interpretability, but still suffer from poor scalability. In this paper, we propose a novel approach to extracting weighted automata with the guidance of a target RNN’s decision and context information. In particular, we identify the patterns of RNN’s step-wise predictive decisions to instruct the formation of automata states. Further, we propose a state composition method to enhance the context-awareness of the extracted model. Our in-depth evaluations on typical RNN tasks, including language model and classification, demonstrate the effectiveness and advantage of our method over the state-of-the-arts. The evaluation results show that our method can achieve accurate approximation of an RNN even on large-scale tasks.
Discipline
OS and Networks | Software Engineering
Research Areas
Software and Cyber-Physical Systems
Publication
Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI 2021), Virtual Conference, February 2-9,
Volume
35
Issue
13
First Page
11699
Last Page
11707
Publisher
AAAI Press
City or Country
Virtual Conference
Citation
ZHANG, Xiyue; DU, Xiaoning; XIE, Xiaofei; MA, Lei; LIU, Yang; and SUN, Meng.
Decision-guided weighted automata extraction from recurrent neural networks. (2021). Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI 2021), Virtual Conference, February 2-9,. 35, (13), 11699-11707.
Available at: https://ink.library.smu.edu.sg/sis_research/7113
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.