ProtoS-ViT: Visual foundation models for sparse self-explainable classifications

Hugues Turbe, Mina Bjelogrlic, Gianmarco Mengaldo, Christian Lovis

Keywords: Explainability, Interpretability, Self-explainable models, Image classification, Transparency, Protoypical part model

TL;DR: This work proposes an extensive sets of metrics to identify flaws in explanations provided by prototypical part models and introduce a novel architecture to address them.

Abstract:

Prototypical networks aim to build intrinsically explainable models based on the
linear summation of concepts. Concepts are coherent entities that we, as humans,
can recognize and associate with a certain object or entity. However, important
challenges remain in the fair evaluation of explanation quality provided by these
models. This work first proposes an extensive set of quantitative and qualitative
metrics which allow to identify drawbacks in current prototypical networks. It
then introduces a novel architecture which provides compact explanations, outperforming
current prototypical models in terms of explanation quality. Overall,
the proposed architecture demonstrates how frozen pre-trained ViT backbones
can be effectively turned into prototypical models for both general and domainspecific
tasks, in our case biomedical image classifiers. Code is available at
https://github.com/hturbe/protosvit.

Track: Main track

Submitted Paper: Yes

Published Paper: No

Submission Number: 71