Explaining an image classifier with a generative model conditioned by uncertainty

We propose to condition a generative model by a given image classifier uncertainty in order to analyze and explain its behavior. Preliminary experiments on synthetic data and a corrupted version of MNIST dataset illustrate the idea.

cs.CV cs.AI eess.IV Computer Science Mathematics Engineering

Source

Paper

References (23)

01Leveraging generative models to characterize the failure conditions of image classifiersAdrien Le Coz, Stéphane Herbin, Faouzi Adjed2024 · AISafety@IJCAI · 1 citation In Library

02Adversarial Counterfactual Visual ExplanationsGuillaume Jeanneret, Loïc Simon, F. Jurie2023 · Computer Vision and Pattern Recognition · 61 citations In Library

03Discovering Bugs in Vision Models using Off-the-shelf Image Generation and CaptioningOlivia Wiles, Isabela Albuquerque, Sven Gowal2022 · arXiv.org · 53 citations In Library

04UQGAN: A Unified Model for Uncertainty Quantification of Deep Classifiers trained via Conditional GANsPhilipp Oberdiek, G. Fink, M. Rottmann2022 · Neural Information Processing Systems · 29 citations In Library

05Generative Adversarial NetworksI. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza et al.2021 · International Conference on Computing Communication and Networking Technologies · 30k citations In Library

06Explaining in Style: Training a GAN to explain a classifier in StyleSpaceOran Lang, Yossi Gandelsman, Michal Yarom et al.2021 · IEEE International Conference on Computer Vision · 185 citations In Library

07Counterfactual Generative NetworksAxel Sauer, Andreas Geiger2021 · International Conference on Learning Representations · 152 citations In Library

08Explainable AI: A Review of Machine Learning Interpretability MethodsPantelis Linardatos, Vasilis Papastefanopoulos, S. Kotsiantis2020 · Entropy · 2.7k citations

09Confidence Estimation via Auxiliary ModelsCharles Corbière, Nicolas Thomé, A. Saporta et al.2020 · IEEE Transactions on Pattern Analysis and Machine Intelligence · 68 citations In Library

10StyleSpace Analysis: Disentangled Controls for StyleGAN Image GenerationZongze Wu, D. Lischinski, Eli Shechtman2020 · Computer Vision and Pattern Recognition · 541 citations In Library

11Denoising Diffusion Probabilistic ModelsJonathan Ho, Ajay Jain, P. Abbeel2020 · Neural Information Processing Systems · 34k citations In Library

12Training Generative Adversarial Networks with Limited DataTero Karras, M. Aittala, Janne Hellsten et al.2020 · Neural Information Processing Systems · 2.3k citations In Library

Scroll for more · 11 remaining

Uncertainty-Aware Deep Classifiers Using Generative ModelsMurat Şensoy, Lance Kaplan, Federico Cerutti et al.2020 · Proceedings of the AAAI Conference on Artificial Intelligence In Library

Leveraging conditional generative models in a general explanation framework of classifier decisionsMartin Charachon, Paul-Henry Cournède, Céline Hudelot et al.2022 · Future Generation Computer Systems In Library

Leveraging Conditional Generative Models in a General Explanation Framework of Classifier DecisionsMartin Charachon, Paul-Henry Cournède, Céline Hudelot et al.2021 · arXiv (Cornell University) · 6 citations In Library

Explaining image classifiers by removing input features using generative modelsChirag Agarwal, D. Schonfeld, Anh Totti Nguyen2019 · arXiv.org · 16 citations In Library

UQGAN: A Unified Model for Uncertainty Quantification of Deep Classifiers trained via Conditional GANsPhilipp Oberdiek, Gernot A. Fink, Matthias Rottmann2022 · arXiv (Cornell University) · 29 citations In Library

Explaining Image Classifiers by Counterfactual GenerationC. Chang, Elliot Creager, A. Goldenberg et al.2018 · International Conference on Learning Representations · 298 citations In Library

Leveraging generative models to characterize the failure conditions of image classifiersAdrien Le Coz, Stéphane Herbin, Faouzi Adjed2024 · arXiv (Cornell University) · 1 citation In Library

Ctrl-U: Robust Conditional Image Generation via Uncertainty-aware Reward ModelingGuiyu Zhang, Huan-ang Gao, Zijian Jiang et al.2024 · International Conference on Learning Representations · 17 citations In Library

Explaining Image Classifiers Using Contrastive Counterfactuals in Generative Latent SpacesKamran Alipour, Aditya Lahiri, Ehsan Adeli et al.2022 · arXiv (Cornell University) In Library

Patents near this paper

SYSTEM AND METHOD FOR PERFORMING CLASSIFICATION THROUGH GENERATIVE MODELS OF FEATURES OCCURRING IN AN IMAGEUS 8,000,538 · PALO ALTO RESEARCH CENTER, INC. · 2011 Systems and Methods for Conditional Generative ModelsUS 11,797,864 · FOTONATION LIMITED · 2023 CLASSIFYING IMAGES UTILIZING GENERATIVE-DISCRIMINATIVE FEATURE REPRESENTATIONSUS 11,709,915 · ADOBE INC. · 2023 Generative Adversarial Network Medical Image Generation for Training of a ClassifierUS 10,592,779 · INTERNATIONAL BUSINESS MACHINES CORPORATION · 2020 Generative Adversarial Network Medical Image Generation for Training of a ClassifierUS 11,645,833 · INTERNATIONAL BUSINESS MACHINES CORPORATION · 2023

Source of record

arXiv

Explaining an image classifier with a generative model conditioned by uncertainty

Paper

References (23)

Similar papers

Patents near this paper

Welcome.

Patents near this paper