General Information
SSD: ING-INF/05
CFU: 6
Lecturer: Eliana Pastor
Teaching Staff: Elena Baralis, Gabriele Ciravegna, Salvatore Greco, Eleonora Poeta
Schedule
- Monday, 8:30-10:00 – classroom 12I
- Thursday, 8:30-10:00 – classroom 7D
- Friday, 8:30-10:00 – classroom 11I
Teaching material
- Course introduction (slides)
- Trustworthy AI: definition and motivations (slides)
- Explainable AI: taxonomy (slides)
- Pre-modeling explainability (slides)
- In-modeling explainability (slides)*
- Post-hoc model agnostic – global (slides)
- Post-hoc model agnostic – local surrogate models (slides)
- Post-hoc model agnostic – local – explaining by removing/perturbing (slides)
- Post-hoc (model agnostic) – local – gradient-based explanation methods (slides)
- Concept-based Explainable AI – part I (slides)
- Concept-based Explainable AI – part II (slides)
- Introduction to NLP (slides)*
- Evaluation of explanations (slides)
- Attention-based Explainability (slides)
- Adversarial Attacks (slides)
- Counterfactual explanations (slides)
Laboratory material
- Lab 0.1: Machine Learning pipeline with Pandas and Scikit-learn (zip) (zip_solution) (pdf_solution)
- Lab 0.2: Introduction to PyTorch with Deep Learning (zip) (zip_solution) (pdf_solution)
- Lab 1: Interpretable by design (zip) (zip_solution) (pdf_solution)
- Lab 2: Post-hoc global explanation methods (zip) (zip_solution)
- Lab 3a: Post-hoc Local explanation methods – LIME (zip) (zip_solution) (pdf_solution)
- Lab 3b: Post-hoc Local explanation methods – SHAP (zip) (zip_solution) (pdf_solution)
- Lab 4a: Image data Analysis & Saliency Maps (zip) (zip_solution) (pdf_solution)
- Lab 4b: XAI for Image data (zip) (zip_solution) (pdf_solution)
- Lab 5: Concept-based XAI (zip) (zip_solution) (pdf_solution_CBM) (pdf_solution_CRAFT)
- Lab 6a: NLP Classifier Training with HuggingFace (zip_solution) (pdf_solution) (data_zip)
- Lab 6b: Explainable NLP (zip) (zip_solution) (pdf_solution) (data_zip)
- Lab 7a, 7b: Evaluating Explanations (zip_a, zip_b) (zip solution)
- Lab 8: Transformer and Attention-based Explainability (old_zip), (zip), (zip_solution), (pdf_solution)
- Lab 9: Adversarial Attacks (zip) (zip_solution)
- Lab 10: Counterfactual explanations (zip) (zip solution)
The material is also available in Github.
Project
- Presentation projects (link)
- Template project Latex (link)
- We suggest using Overleaf
- List projects (pdf, separate)
Exam
Exam sample (pdf)