Does a Neural Network Really Encode Symbolic Concepts?

Mingjie Li, Quanshi Zhang
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:20452-20469, 2023.

Abstract

Recently, a series of studies have tried to extract interactions between input variables modeled by a DNN and define such interactions as concepts encoded by the DNN. However, strictly speaking, there still lacks a solid guarantee whether such interactions indeed represent meaningful concepts. Therefore, in this paper, we examine the trustworthiness of interaction concepts from four perspectives. Extensive empirical studies have verified that a well-trained DNN usually encodes sparse, transferable, and discriminative concepts, which is partially aligned with human intuition. The code is released at https://github.com/sjtu-xai-lab/interaction-concept.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-li23at, title = {Does a Neural Network Really Encode Symbolic Concepts?}, author = {Li, Mingjie and Zhang, Quanshi}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {20452--20469}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/li23at/li23at.pdf}, url = {https://proceedings.mlr.press/v202/li23at.html}, abstract = {Recently, a series of studies have tried to extract interactions between input variables modeled by a DNN and define such interactions as concepts encoded by the DNN. However, strictly speaking, there still lacks a solid guarantee whether such interactions indeed represent meaningful concepts. Therefore, in this paper, we examine the trustworthiness of interaction concepts from four perspectives. Extensive empirical studies have verified that a well-trained DNN usually encodes sparse, transferable, and discriminative concepts, which is partially aligned with human intuition. The code is released at https://github.com/sjtu-xai-lab/interaction-concept.} }
Endnote
%0 Conference Paper %T Does a Neural Network Really Encode Symbolic Concepts? %A Mingjie Li %A Quanshi Zhang %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-li23at %I PMLR %P 20452--20469 %U https://proceedings.mlr.press/v202/li23at.html %V 202 %X Recently, a series of studies have tried to extract interactions between input variables modeled by a DNN and define such interactions as concepts encoded by the DNN. However, strictly speaking, there still lacks a solid guarantee whether such interactions indeed represent meaningful concepts. Therefore, in this paper, we examine the trustworthiness of interaction concepts from four perspectives. Extensive empirical studies have verified that a well-trained DNN usually encodes sparse, transferable, and discriminative concepts, which is partially aligned with human intuition. The code is released at https://github.com/sjtu-xai-lab/interaction-concept.
APA
Li, M. & Zhang, Q.. (2023). Does a Neural Network Really Encode Symbolic Concepts?. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:20452-20469 Available from https://proceedings.mlr.press/v202/li23at.html.

Related Material