DISK-CSV: Distilling Interpretable Semantic Knowledge with a Class Semantic Vector

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 3021–3030 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 3021 DISK-CSV: Distilling Interpretable Semantic Knowledge with a Class Semantic Vector Housam Khalifa Bashier Babiker 1, Mi-Young Kim2, Randy Goebel1 Alberta Machine Intelligence Institute 1 Department of Computing Science, University of Alberta 2 Department of Science, Augustana Faculty, University of Alberta {khalifab, miyoung2, rgoebel}@ulaberta.ca Abstract Neural networks (NN) applied to natural lan- guage processing (NLP) are becoming deeper and more complex, making them increasingly difficult to understand and interpret. Even in applications of limited scope on fixed data, the creation of these complex “black-boxes” cre- ates substantial challenges for debugging, un- derstanding, and generalization. But rapid de- velopment in this field has now lead to building more straightforward and interpretable mod- els. We propose a new technique (DISK- CSV) to distill knowledge concurrently from any neural network architecture for text clas- sification, captured as a lightweight inter- pretable/explainable classifier. Across multi- ple datasets, our approach achieves better per- formance than the target black-box. In ad- dition, our approach provides better explana- tions than existing techniques. 1 Introduction Deep Neural Networks (DNNs) are popular in many applications, including computer vision and natural language processing. However, two major factors still require attention: (1) understanding the classifier’s prediction for the end-users to develop trust in the model, and to enable machine learn- ing engineers to refine it. This is especially true for high-stakes domains such as clinical decision support. There has been attention on models that attempt to make a neural network explainable, for instance, (Sundararajan et al., 2017), and (Ribeiro et al., 2016), which create post-hoc explanations to support explainability. Secondly, (2) artificially high numbers of parameters make the inference time expensive. Figure 1: Our proposed model. Neural networks tend to be deep, with millions of parameters. For example, GPT-2 (Radford et al., 2019) needs over 1.5 billion parameters. As a re- sult, they are compute-intensive, thus making it difficult to deploy in real-world applications. We here propose a model-agnostic interpretable knowl- edge distillation method for neural network text classification. As shown in Figure 1, we learn a class seman- tic vector for each output class, concurrently when training the black box. We then use the seman- tic vectors to create the nearest neighbor classifier (compressed interpretable/explainable classifier ) from the black-box version. Knowledge distillation refers to the process of transferring the implicit knowledge learned by a teacher model to a student model (Liu and Matwin, 2018). Dark knowledge refers to the salient information hidden in the pre- dicted probabilities for all classes, which are more informative than the predicted classes themselves. Our contributions can be summarized as follows: • We propose a knowledge distillation method where dark knowledge can be learned concur- rently by a student model, while building a black-box model. • We propose an interpretable classifier, which provides a user explanation for predicting a single class label. 3022 • We integrate a …

Content truncated to 500 words