[{"command":"insert","method":null,"selector":".course-modal-ajax-placeholder-959636","data":"\u003Cdiv class=\u0022container content-rte\u0022\u003E\n  \u003Cdiv class=\u0022row\u0022\u003E\n    \u003Cdiv class=\u0022box col-md-6 bg-primary text-white\u0022\u003E\n      \u003Ch2 class=\u0022h1 text-left\u0022\u003E  Multimodal Explainable AI (XAI)\n\u003C\/h2\u003E\n                  \u003Cdiv class=\u0022field-spacing\u0022\u003E\n        \u003Cdiv class=\u0022field__label h3 dash\u0022\u003EECTS :\n          5\n        \u003C\/div\u003E\n      \u003C\/div\u003E\n                  \u003Cdiv class=\u0022field-spacing\u0022\u003E\n        \u003Cdiv class=\u0022field__label h3 dash\u0022\u003ESemester:\n          Annualis\u00e9\n        \u003C\/div\u003E\n      \u003C\/div\u003E\n            \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003ED\u00e9tail du volume horaire\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                          \u003Cspan class=\u0022field__label h4 mt-5 \u0022\u003ELecture:\u003C\/span\u003E\n                                      \u003Cspan\u003E10\u003C\/span\u003E\n                      \u003C\/div\u003E\n                                              \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                          \u003Cspan class=\u0022field__label h4 mt-5 \u0022\u003EDirected study:\u003C\/span\u003E\n                                      \u003Cspan\u003E17\u003C\/span\u003E\n                      \u003C\/div\u003E\n                                              \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                          \u003Cspan class=\u0022field__label h4 mt-5 \u0022\u003EProject :\u003C\/span\u003E\n                                      \u003Cspan\u003E2\u003C\/span\u003E\n                      \u003C\/div\u003E\n                                \u003C\/div\u003E\n        \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003ELangue d\u0026#039;enseignement\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003EAnglais\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n        \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003EEnseignement \u00e0 distance\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003Enon\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n      \u003C\/div\u003E\n    \u003Cdiv class=\u0022box col-md-6\u0022\u003E\n                    \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003EProgramme \/ plan \/ contenus\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003E\u003Cp\u003EWeek 1: Introduction to Explainable AI (XAI)\u003Cbr\u003E\n\u25cf Overview of XAI:\u003Cbr\u003E\n\u25cb What is Explainable AI?\u003Cbr\u003E\n\u25cb Importance and challenges of interpretability\u003Cbr\u003E\n\u25cb Real-world applications of XAI (e.g., healthcare, finance, autonomous systems)\u003Cbr\u003E\n\u25cb Introduction to different types of explanations in AI\u003Cbr\u003E\n\u25cf LIME \u0026amp; SHAP (post-hoc explanation):\u003Cbr\u003E\n\u25cb Understanding Local Interpretable Model-agnostic Explanations (LIME) [2]\u003Cbr\u003E\n\u25cb Introduction to SHAP (SHapley Additive exPlanations) [3] and Kernel SHAP\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Implementing LIME and SHAP on vision tasks (TP on vision)\u003C\/p\u003E\n\u003Cp\u003EWeek 2 Introduction to Explainable AI (XAI)\u003Cbr\u003E\n\u25cf Attribution Methods (CAM \u0026amp; Grad-CAM \u0026amp; Gradient based strategy post-hoc explanation) :\u003Cbr\u003E\n\u25cb Class Activation Mapping (CAM)\u003Cbr\u003E\n\u25cb Gradient-weighted Class Activation Mapping (Grad-CAM)[1]\u003Cbr\u003E\n\u25cb Application to convolutional neural networks (CNNs)\u003Cbr\u003E\n\u25cb Integrated gradient\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Using CAM and Grad-CAM to interpret vision models (TP on vision)\u003C\/p\u003E\n\u003Cp\u003EWeek 3: Variance-based Sensitivity Analysis\u003Cbr\u003E\n\u25cf Variance-based Sensitivity Analysis (post-hoc explanation) :\u003Cbr\u003E\n\u25cb Overview of sensitivity analysis methods\u003Cbr\u003E\n\u25cb Mathematical foundations and applications\u003Cbr\u003E\n\u25cb Review of key articles on variance-based methods for XAI [4][5]\u003Cbr\u003E\n\u25cf Hands-on Tutorial :Applying variance-based sensitivity analysis on audio models (TP on audio)\u003C\/p\u003E\n\u003Cp\u003EWeek 4: Counterfactual Explanations\u003Cbr\u003E\n\u25cf Understanding Counterfactuals :\u003Cbr\u003E\n\u25cb Definition and theory of counterfactual explanations [6]\u003Cbr\u003E\n\u25cb Application in AI models for decision-making transparency\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Counterfactual explanations for audio-based AI systems (TP on audio)\u003C\/p\u003E\n\u003Cp\u003EWeek 5: Concept Bottleneck Models\u003Cbr\u003E\n\u25cf Concept Bottleneck Models :\u003Cbr\u003E\n\u25cb Overview of Concept Bottleneck Models and their role in improving interpretability [8,9]\u003Cbr\u003E\n\u25cb Using intermediate concepts to explain final predictions\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Implementing and evaluating Concept Bottleneck Models for vision-based tasks (TP on vision)\u003C\/p\u003E\n\u003Cp\u003EWeek 6: Concept Activation Vectors (CAVs)\u003Cbr\u003E\n\u25cf Concept Activation Vectors :\u003Cbr\u003E\n\u25cb Introduction to CAVs for interpreting deep learning models [10]\u003Cbr\u003E\n\u25cb How CAVs connect human-understandable concepts to neural network activations\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Using CAVs to interpret vision-language models (TP on vision-NLP)\u003C\/p\u003E\n\u003Cp\u003EWeek 7: Prototype Networks\u003Cbr\u003E\n\u25cf Prototype Networks :\u003Cbr\u003E\n\u25cb Understanding Prototype Networks and how they improve explainability by leveraging representative prototypes for predictions [11]\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Building and testing Prototype Networks for audio-related tasks (TP on audio)\u003C\/p\u003E\n\u003Cp\u003EWeek 8: Chain of Thought Reasoning\u003Cbr\u003E\n\u25cf Chain of Thought Reasoning :\u003Cbr\u003E\n\u25cb How models can simulate human-like reasoning processes through chains of thought [12]\u003Cbr\u003E\n\u25cb Importance of step-by-step explanations in complex decision-making systems\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Applying Chain of Thought reasoning in NLP models (TP on NLP)\u003C\/p\u003E\n\u003Cp\u003EWeek 9: Knowledge Graphs in XAI\u003Cbr\u003E\n\u25cf Knowledge Graphs :\u003Cbr\u003E\n\u25cb Introduction to knowledge graphs and their use in enhancing model interpretability [13]\u003Cbr\u003E\n\u25cb Role of knowledge graphs in NLP and AI systems for structured explanations\u003Cbr\u003E\n\u25cf Hands-on Tutorial : Integrating knowledge graphs into NLP models (TP on NLP)\u003C\/p\u003E\n\u003Cp\u003EWeek 10: Final Project and Exam\u003Cbr\u003E\n\u25cf Project and Exam:\u003Cbr\u003E\n\u25cb Final exam (1h)\u003Cbr\u003E\n\u25cb Students work on a final project to apply multiple XAI techniques across a multimodal dataset (2h)\u003C\/p\u003E\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n        \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003EObjectifs d\u0026#039;apprentissage\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003E\u003Cp\u003EThis course explores Explainable Artificial Intelligence (XAI), a crucial subfield of machine learning dedicated to enhancing the transparency of complex models. While modern AI systems\u2014particularly Deep Neural Networks (DNNs) and Foundation Models achieve state-of-the-art performance, their black-box nature makes it challenging to understand the reasoning behind their predictions. This lack of interpretability raises concerns about trust, accountability, and the ability to extract meaningful insights from these models.\u003Cbr\u003E\nThe course examines two key perspectives in XAI:\u003Cbr\u003E\n1. The argument for using inherently interpretable models in high-stakes domains such as healthcare and finance.\u003Cbr\u003E\n2. The development of post hoc explanation techniques that provide insight into complex models after training.\u003Cbr\u003E\nStudents will engage with a variety of state-of-the-art XAI methods across multiple modalities, including computer vision, audio processing, and natural language processing (NLP). Topics covered include attribution techniques, sensitivity analysis, Concept Bottleneck Models, Concept Activation Vectors (CAVs), and Counterfactual Explanations. Through hands-on exercises, students will gain practical experience applying XAI techniques, equipping them to enhance transparency and interpretability across diverse AI applications.\u003C\/p\u003E\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n                                \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003EBibliographie\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003E\u003Cp\u003ERequired Reading and Resources:\u003Cbr\u003E\n1. https:\/\/github.com\/jacobgil\/pytorch-grad-cam\u003Cbr\u003E\n2. Ribeiro, M. T., Singh, S., \u0026amp; Guestrin, C. (2016). \u00ab Why should I trust you? \u00bb Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.\u003Cbr\u003E\n3. Lundberg, S. M., \u0026amp; Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems (NeurIPS).\u003Cbr\u003E\n4. Sobol, Ilya M. \u00ab Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates. \u00bb Mathematics and computers in simulation 55.1-3 (2001): 271-280.\u003Cbr\u003E\n5. Fel, T., Cad\u00e8ne, R., Chalvidal, M., Cord, M., Vigouroux, D., \u0026amp; Serre, T. (2021). Look at the variance! efficient black-box explanations with sobol-based sensitivity analysis. Advances in neural information processing systems, 34, 26005-26014.\u003Cbr\u003E\n6. Mothilal, Ramaravind K., Amit Sharma, and Chenhao Tan. \u00ab Explaining machine learning classifiers through diverse counterfactual explanations. \u00bb Proceedings of the 2020 conference on fairness, accountability, and transparency. 2020.\u003Cbr\u003E\n7. Agerri, R., \u0026amp; Garc\u00eda-Serrano, A. (2021). \u00ab Multimodal deep learning: A review of the state of the art. \u00bb arXiv:2111.04138.\u003Cbr\u003E\n8. Koh, Pang Wei, et al. \u00ab Concept bottleneck models. \u00bb International conference on machine learning. PMLR, 2020.\u003Cbr\u003E\n9. Kazmierczak, R., Berthier, E., Frehse, G., \u0026amp; Franchi, G. (2023). CLIP-QDA: An explainable concept bottleneck model. arXiv preprint arXiv:2312.00110.\u003Cbr\u003E\n10. Parekh, J., Khayatan, P., Shukor, M., Newson, A., \u0026amp; Cord, M. (2024). A concept-based explainability framework for large multimodal models. Advances in Neural Information Processing Systems, 37, 135783-135818.\u003Cbr\u003E\n11. Donnelly, Jon, Alina Jade Barnett, and Chaofan Chen. \u00ab Deformable protopnet: An interpretable image classifier using deformable prototypes. \u00bb Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition. 2022.\u003Cbr\u003E\n12. Saparov, Abulhair, and He He. \u00ab Language models are greedy reasoners: A systematic formal analysis of chain-of-thought. \u00bb arXiv preprint arXiv:2210.01240 (2022).\u003Cbr\u003E\n13. Tiddi, Ilaria, and Stefan Schlobach. \u00ab Knowledge graphs as tools for explainable machine learning: A survey. \u00bb Artificial Intelligence 302 (2022): 103627.\u003Cbr\u003E\n14. https:\/\/christophm.github.io\/interpretable-ml-book\/\u003C\/p\u003E\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n        \u003Cdiv class=\u0022field-spacing\u0022\u003E\n      \u003Cdiv class=\u0022field__label h3 dash\u0022\u003ENature de l\u0026#039;\u00e9valuation\u003C\/div\u003E\n                        \u003Cdiv class=\u0022mt-5 text-justify\u0022\u003E\n                                      \u003Cspan\u003ED\u00e9tail renseign\u00e9 au niveau \u00c9l\u00e9ment Constitutif\u003C\/span\u003E\n                      \u003C\/div\u003E\n                  \u003C\/div\u003E\n      \u003C\/div\u003E\n  \u003C\/div\u003E\n\u003C\/div\u003E\n","settings":null}]