3D Driven Explainable AI

Event details
Date | 26.06.2023 |
Hour | 16:00 › 18:00 |
Speaker | Yann Bouquet |
Location | |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Pascal Frossard
Thesis advisor: Prof. Mathieu Salzmann
Co-examiner: Prof. Amir Zamir
Abstract
Explainable AI (XAI) methods are being explored to address the lack of transparency and interpretability in neural networks. This research proposes using 3D parameters to generate precise post hoc explanations, improving user control and assessing the robustness of deep learning models. By incorporating generative models for 3D data and employing techniques like signal decompositions, the goal is to enhance understanding of computer vision models. The study aims to leverage modular infrastructure and generative models in the 3D domain to identify semantic failure modes and improve control over explanations. We aim our contribution to participate in the development of counterfactual explanations that can go beyond the limited attributes and datasets taken into account in current methods.
Background papers
Leclerc Guillaume, Hadi Salman, Andrew Ilyas, Sai Vemprala, Logan Engstrom, Vibhav Vineet, Kai Xiao, et al. “3DB: A Framework for Debugging Computer Vision Models.” In _Advances in Neural Information Processing Systems_, edited by S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, 35:8498–8511. Curran Associates, Inc., 2022.
https://proceedings.neurips.cc/paper_files/paper/2022/file/3848bc3112429079af85dedb7d369ef4-Paper-Conference.pdf
Jeanneret Guillaume, Simon Loic and Frederic Jurie. "Adversarial Counterfactual Visual Explanations." In _2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023
https://arxiv.org/pdf/2303.09962.pdf
Hui Ka-Hei, Ruihui Li, Jingyu Hu, and Chi-Wing Fu. “Neural Template: Topology-Aware Reconstruction and Disentangled Generation of 3D Meshes.” In _2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)_, 18551–61. New Orleans, LA, USA: IEEE, 2022.
https://openaccess.thecvf.com/content/CVPR2022/papers/Hui_Neural_Template_Topology-Aware_Reconstruction_and_Disentangled_Generation_of_3D_Meshes_CVPR_2022_paper.pdf
Exam president: Prof. Pascal Frossard
Thesis advisor: Prof. Mathieu Salzmann
Co-examiner: Prof. Amir Zamir
Abstract
Explainable AI (XAI) methods are being explored to address the lack of transparency and interpretability in neural networks. This research proposes using 3D parameters to generate precise post hoc explanations, improving user control and assessing the robustness of deep learning models. By incorporating generative models for 3D data and employing techniques like signal decompositions, the goal is to enhance understanding of computer vision models. The study aims to leverage modular infrastructure and generative models in the 3D domain to identify semantic failure modes and improve control over explanations. We aim our contribution to participate in the development of counterfactual explanations that can go beyond the limited attributes and datasets taken into account in current methods.
Background papers
Leclerc Guillaume, Hadi Salman, Andrew Ilyas, Sai Vemprala, Logan Engstrom, Vibhav Vineet, Kai Xiao, et al. “3DB: A Framework for Debugging Computer Vision Models.” In _Advances in Neural Information Processing Systems_, edited by S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, 35:8498–8511. Curran Associates, Inc., 2022.
https://proceedings.neurips.cc/paper_files/paper/2022/file/3848bc3112429079af85dedb7d369ef4-Paper-Conference.pdf
Jeanneret Guillaume, Simon Loic and Frederic Jurie. "Adversarial Counterfactual Visual Explanations." In _2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023
https://arxiv.org/pdf/2303.09962.pdf
Hui Ka-Hei, Ruihui Li, Jingyu Hu, and Chi-Wing Fu. “Neural Template: Topology-Aware Reconstruction and Disentangled Generation of 3D Meshes.” In _2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)_, 18551–61. New Orleans, LA, USA: IEEE, 2022.
https://openaccess.thecvf.com/content/CVPR2022/papers/Hui_Neural_Template_Topology-Aware_Reconstruction_and_Disentangled_Generation_of_3D_Meshes_CVPR_2022_paper.pdf
Practical information
- General public
- Free
Contact
- edic@epfl.ch