Causethical Machine Learning 2.0

Now in its second iteration, this seminar explores the intersection of causality with robustness, interpretability, and fairness in machine learning, exploring how causal reasoning can help build more reliable, explainable, and fair AI systems. As AI increasingly influences critical domains such as healthcare, finance, and policy, ensuring that models make sound, transparent, and justifiable decisions is more important than ever. This seminar examines how causal methods provide a principled approach to addressing key challenges in AI, moving beyond correlations to uncover true cause-and-effect relationships.

Structured into three core themes: Robustness, Interpretability, and Fairness, the seminar engages participants in an in-depth exploration of causal theory, methodological innovations, and real-world applications. Through a carefully curated selection of foundational and state-of-the-art research, students will analyze how causality enhances model reliability, improves decision transparency, and mitigates bias in AI systems.

The three main blocks are organized according to these core themes:

• Block I: Fairness and Causality
• Block II: Robustness and Causality
• Block III: Interpretability and Causality

Each session will feature discussions on recent research, student-led paper presentations, and critical engagement with open problems in the field. By blending theoretical insights with practical motivation, this seminar provides students with a deep, structured understanding of how causal methods contribute to the development of Responsible and Ethical AI.

Date and time: Weekly. Tuesday 12:30 - 14:00.

Location:  SR 106 E1.1

Number of Students: 12

Attendance: Regular attendance is mandatory. Students may miss up to two sessions, for which they must be excused. In case of absence, students must still ensure they provide all deliverables by the due dates.

Seminar Schedule Overview

Timeline of Causethical seminar 2.0v

Dates.	Block	Content
22/04/2025		Introduction to Seminar, Causality, Responsible AI
29/04/2025	Block I	Introduction to Block I: Fairness and Causality
6/05/2025	Block I	Student Presentation + QnA x2
13/05/2025	Block I	Panel Discussion
20/05/2025	Block I	Guest Lecture
27/05/2025	Block II	Introduction to Block II: Robustness and Causality
03/06/2025	Block II	Student Presentation Paper #1 and #2
10/06/2025	Block II	Student Presentation Paper #3 and #4
17/06/2025	Block II	Panel Discussion
24/06/2025	Block III	Introduction to Block III: Interpretability and Causality
01/07/2025	Block III	Student Presentation Papers #1 and #2
08/07/2025	Block III	Student Presentation Papers #3 and #4
15/07/2025	Block III	Panel Discussion

Pool of papers : This selection of state-of-the-art papers represents what we believe are valuable resources for discussing the topics covered in this seminar.

Please select the block preference by 21/04/2025. The final decision of the block will be decided by TAs taking into account the preferences of the student. 

Presentation: The papers of the presentation will be given by TAs however you are free to supplement your presentation with more relevant articles. We always encourage addition of recent literature. 
 

Block I: Fairness and Causality

• Avoiding Discrimination through Causal Reasoning
• Counterfactual Fairness
• Fairness-Accuracy Trade-Offs: A Causal Perspective
• Counterfactual Fairness Is Basically Demographic Parity
• Promises and Challenges of Causality for Ethical Machine Learning
• Counterfactual Fairness Is Not Demographic Parity, and Other Observations
• Causal Conceptions of Fairness and their Consequences

Block II: Robustness and Causality

• [Paper #1 7073409]: Representation Learning via Invariant Causal Mechanisms 
• [Paper #2 7070603]: Learning Causally Disentangled Representations via the Principle of Independent Causal Mechanisms 
• [Paper #3 7047009]: Causality-Inspired Representation Learning for Domain Generalization 
• [Paper #4 7068863]:  Invariant Causal Representation Learning for Out-of-Distribution Generalization 
• Learning causal representations for robust domain adaptation
• Robust Causal Graph Representation Learning against Confounding Effects

Block III: Interpretability and Causality

• [Paper #1 7039983] Causally Reliable Concept Bottleneck Models
• [Paper #2 7072540] Causal Concept Graph Models: Beyond Causal Opacity in Deep Learning
• [Paper #3 7072402] Explaining Classifiers with Causal Concept Effect 
• [Paper #4 7070990] Interpretability is in the mind of the beholder: A causal framework for human-interpretable representation learning
• From Causal to Concept-Based Representation Learning