This Looks Like That, Because.. Explaining Prototypes for Interpretable Image Recognition

Meike Nauta*, Annemarie Jutte, Jesper Provoost, Christin Seifert

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

18 Citations (Scopus)
136 Downloads (Pure)

Abstract

Image recognition with prototypes is considered an interpretable alternative for black box deep learning models. Classification depends on the extent to which a test image “looks like” a prototype. However, perceptual similarity for humans can be different from the similarity learned by the classification model. Hence, only visualising prototypes can be insufficient for a user to understand what a prototype exactly represents, and why the model considers a prototype and an image to be similar. We address this ambiguity and argue that prototypes should be explained. We improve interpretability by automatically enhancing visual prototypes with quantitative information about visual characteristics deemed important by the classification model. Specifically, our method clarifies the meaning of a prototype by quantifying the influence of colour hue, shape, texture, contrast and saturation and can generate both global and local explanations. Because of the generality of our approach, it can improve the interpretability of any similarity-based method for prototypical image recognition. In our experiments, we apply our method to the existing Prototypical Part Network (ProtoPNet). Our analysis confirms that the global explanations are generalisable, and often correspond to the visually perceptible properties of a prototype. Our explanations are especially relevant for prototypes which might have been interpreted incorrectly otherwise. By explaining such ‘misleading’ prototypes, we improve the interpretability and simulatability of a prototype-based classification model. We also use our method to check whether visually similar prototypes have similar explanations, and are able to discover redundancy. Code is available at https://github.com/M-Nauta/Explaining_Prototypes.

Original languageEnglish
Title of host publicationMachine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2021, Proceedings
EditorsMichael Kamp, Michael Kamp, Irena Koprinska, Adrien Bibal, Tassadit Bouadi, Benoît Frénay, Luis Galárraga, José Oramas, Linara Adilova, Yamuna Krishnamurthy, Bo Kang, Christine Largeron, Jefrey Lijffijt, Tiphaine Viard, Pascal Welke, Massimiliano Ruocco, Erlend Aune, Claudio Gallicchio, Gregor Schiele, Franz Pernkopf, Michaela Blott, Holger Fröning, Günther Schindler, Riccardo Guidotti, Anna Monreale, Salvatore Rinzivillo, Przemyslaw Biecek, Eirini Ntoutsi, Mykola Pechenizkiy, Bodo Rosenhahn, Christopher Buckley, Daniela Cialfi, Pablo Lanillos, Maxwell Ramstead, Tim Verbelen, Pedro M. Ferreira, Giuseppina Andresini, Donato Malerba, Ibéria Medeiros, Philippe Fournier-Viger, M. Saqib Nawaz, Sebastian Ventura, Meng Sun, Min Zhou, Valerio Bitetta, Ilaria Bordino, Andrea Ferretti, Francesco Gullo, Giovanni Ponti, Lorenzo Severini, Rita Ribeiro, João Gama, Ricard Gavaldà, Lee Cooper, Naghmeh Ghazaleh, Jonas Richiardi, Damian Roqueiro, Diego Saldana Miranda, Konstantinos Sechidis, Guilherme Graça
PublisherSpringer
Pages441-456
Number of pages16
Volume1524
ISBN (Print)9783030937355
DOIs
Publication statusPublished - 1 Jan 2022
Event21st European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021 - Virtual, Online
Duration: 13 Sept 202117 Sept 2021
Conference number: 21

Publication series

NameCommunications in Computer and Information Science
Volume1524 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference21st European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021
Abbreviated titleECML PKDD 2021
CityVirtual, Online
Period13/09/2117/09/21

Keywords

  • 22/1 OA procedure

Fingerprint

Dive into the research topics of 'This Looks Like That, Because.. Explaining Prototypes for Interpretable Image Recognition'. Together they form a unique fingerprint.

Cite this