Towards Robust Metrics For Concept Representation Evaluation

Conference Object
Change log
Zarlenga, ME 
Barbiero, P 
Shams, Z 
Kazhdan, D 
Bhatt, U 

Recent work on interpretability has focused on concept-based explanations, where deep learning models are explained in terms of high-level units of information, referred to as con- cepts. Concept learning models, however, have been shown to be prone to encoding impurities in their representations, failing to fully capture meaningful features of their inputs. While concept learning lacks metrics to measure such phe- nomena, the field of disentanglement learning has explored the related notion of underlying factors of variation in the data, with plenty of metrics to measure the purity of such factors. In this paper, we show that such metrics are not ap- propriate for concept learning and propose novel metrics for evaluating the purity of concept representations in both ap- proaches. We show the advantage of these metrics over exist- ing ones and demonstrate their utility in evaluating the robust- ness of concept representations and interventions performed on them. In addition, we show their utility for benchmark- ing state-of-the-art methods from both families and find that, contrary to common assumptions, supervision alone may not be sufficient for pure concept representations.

Journal Title
Proceedings of the 37th AAAI Conference on Artificial Intelligence, AAAI 2023
Conference Name
Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023
Journal ISSN
Volume Title
Publisher DOI
Publisher URL