PeXAI

Time: Mondays 16:30, Thursday 14:15

1. 10.04 (Thursday)

   • organisational matters

   • assignment of presentation topics

   • brief introduction to XAI

2. 14.04 (Monday)

   • presentation: Saliency Maps - Vanilla Gradient  & GradCAM & DeepLift (by Mikołaj H.)

3. 24.04  (Thursday)

   • Project - Topic Presentations

4. 28.04  (Monday)

   • presentation: Variable Importance (Linear Regression, Random Forest, xGBoost) (by Emre B., Jakub M.)

   • presentation: Feature Visualization, Importance & Sensitivity Analysis for Images (by Jan B., Dominik B.)

5. 08.05 (Thursday)

   • presentation: Inherently Interpretable Models & Hybrid Approaches (by Natalia P., Martyna F.)

   • presentation: Shapley-Values & SHAP (by Dawid K., TBA)

   • prepare a PoC for your project (basic analysis of the data, simple preprocessing, build a black-box model, evalutae the model, add any explanation method to it, make a short summary of the results)

6. 12.05  (Monday)

   • presentation: LIME (by Agnieszka G, Zuzanna K.)

   • presentation: EDA (by Mateusz G., TBA)

7. 15.05 (Thursday)

8. 19.05  (Monday)

9. 22.05 (Thursday)

10. 26.05  (Monday)

    • Project - Milestone Presentations

11. 29.05 (Thursday)

12. 02.06  (Monday)

13. 05.06 (Thursday)

14. 09.06  (Monday)

15. 12.06 (Thursday)

    • Project - Final Presentations

1. Attendance is mandatory. A maximum of 3 absences are allowed (you do not need to state the reason).

2. Each student must contribute to two presentations (30-40 min each, includes implementing examples): one on basic topics (with implementation of the method in a predefined interface) and one directly related to the realized project. In case of insufficient quality, fixes to the programming part of the task may be required. 

3. Each student is required to implement the XAI module into their own ML project (in groups of 2-3 people). Technical quality, suitability and variety of methods, complexity and advancement  of the project will determine the grade awarded. 

4. You can get 25 points from the course:

   • 5 p. for the first presentation (at least 3 p. to pass the course)

   • 8 p. for the second presentation (at least 4 p. to pass the course)

   • 12 p. from the project

5. To pass the course, you need to obtain the required number of points from the presentations and meet the presence criteria. Then, the thresholds for each grade are as follows:

   • 12.5 p. -- 3.0

   • 15 p. -- 3.5

   • 17.5 p. -- 4.0

   • 20 p. -- 4.5

   • 22.5 p. -- 5.0

6. All presented materials (slides, code examples) should be provided to the tutor so that they can be made available to other participants in the class. Codes should be put into https://github.com/klaudiabalcer/PeXAI2025 (please create a PR to main when your code will be ready) and presentations should be uploaded using the form  https://forms.gle/jjdDPFprWvFFY2AJ8 (please remember to specify all team members).

1. Basic presentations: 

   • TBA

2. Advanced Presentations: 

   • TBA

3. Basic Literature:

   • https://www.sciencedirect.com/science/article/pii/S1566253523001148

   • https://ema.drwhy.ai/

   • https://christophm.github.io/interpretable-ml-book/ 

4. Advanced Literature Reviews:

   • https://xainlp2020.github.io/xainlp/table

   • https://github.com/hbaniecki/adversarial-explainable-ai

   • https://github.com/SinaMohseni/Awesome-XAI-Evaluation

   • https://github.com/wangyongjie-ntu/Awesome-explainable-AI

   • https://github.com/pbiecek/xai_resources

   • https://github.com/lopusz/awesome-interpretable-machine-learning

   • https://github.com/rehmanzafar/xai-iml-sota

   • https://github.com/kevinmcareavey/chai-xai

   • https://github.com/feifeife/All-about-XAI

   • https://github.com/samzabdiel/XAI

   • https://github.com/AstraZeneca/awesome-explainable-graph-reasoning

   • https://github.com/anguyen8/XAI-papers

   • https://github.com/JHoelli/Awesome-Time-Series-Explainability?tab=readme-ov-file

5. Advanced Literature - selected papers:

   • Images:

     • https://papers.nips.cc/paper_files/paper/2019/hash/adf7ee2dcf142b0e11888e72b43fcb75-Abstract.html

     • https://arxiv.org/abs/2011.02863

     • https://arxiv.org/abs/2012.02046

     • https://dl.acm.org/doi/10.1145/3580305.3599290 

     • https://arxiv.org/abs/2110.15355

     • https://arxiv.org/abs/2108.11761

     • https://proceedings.neurips.cc/paper_files/paper/2022/hash/025f7165a452e7d0b57f1397fed3b0fd-Abstract-Conference.html

     • https://arxiv.org/abs/2208.03608

     • https://openaccess.thecvf.com/content/ICCV2021/html/Lee_LFI-CAM_Learning_Feature_Importance_for_Better_Visual_Explanation_ICCV_2021_paper.html

   • Tabular Data:

     • https://github.com/peymanrasouli/EXPLAN

     • https://arxiv.org/abs/1911.02042

     • https://arxiv.org/abs/1909.09369

     • https://arxiv.org/abs/1905.04610

     • https://dl.acm.org/doi/abs/10.1145/3412815.3416893

     • https://arxiv.org/abs/2202.00858

     • https://arxiv.org/abs/2104.02297

   • Graph-based Models:

     • https://arxiv.org/abs/2006.02587

     • https://arxiv.org/abs/2203.15209

     • https://arxiv.org/abs/1903.03894

     • https://proceedings.neurips.cc/paper_files/paper/2022/hash/a69d7f3a1340d55c720e572742439eaf-Abstract-Conference.html

     • https://arxiv.org/abs/2102.03322

   • NLP & LLMs:

     • https://arxiv.org/pdf/2010.00577

     • https://arxiv.org/abs/2202.07304

     • https://openreview.net/forum?id=H1z-PsR5KX

     • https://arxiv.org/abs/2010.02695

     • https://arxiv.org/abs/2212.10534

     • https://arxiv.org/abs/1906.02715

   • Audio:

     • https://openreview.net/pdf?id=FhuM-kk8Pbk

   • Temporal (Sequential, RS, Time Series):

     • https://arxiv.org/abs/2107.02776

     • https://dl.acm.org/doi/10.1145/3292500.3330989

     • https://arxiv.org/abs/2408.03636 

     • https://proceedings.mlr.press/v151/sivill22a.html

     • https://arxiv.org/abs/2003.02821

     • https://www.bu.edu/peaclab/files/2021/05/CoMTE___ICAPAI.pdf

     • https://arxiv.org/abs/2106.05303

     • https://arxiv.org/abs/2009.04796