| MATH80630 | Lectures | Assignments | Project | Office hour

Trustworthy Machine Learning (MATH80630): Fall 2022

Paper assingmeents

The paper assingments will be worth 30% of your final grade. Each student needs to submit a report for the weekly assigned readings (15%). Also, each student needs to present at-least one paper and lead a panel discussion (15%).

You can use this latex template to submit your paper reviews.

Grading Scheme

Paper Review (15%)

• Paper summeries: 3%
• Paper strengths: 3%
• Paper weakness: 3%
• Research questions: 3%
• Panel questions: 3%

Paper Presentation (15%)

• Clarity of presentation: 3%
• Slide quality: 3%
• Correctness: 3%
• Answers to questions: 3%
• Panel: 3%

Assingment Schedule

---

1- Week 1 (September 2): Class introduction and machine learning review

• Paper assignments:
  • No paper assignments!

---

2- Week 2 (September 9): Statistical Fairness Definitions in Machine Learning

• Paper assignments:
  • Counterfactual fairness
  • On the (im)possibility of fairness

---

3- Week 3 (September 16): Causal Fairness Definitions in Machine Learning

• Paper assignments:
  • Men Also Like Shopping: Reducing Gender Bias Amplification Using Corpus Level Constraints
  • Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations

---

4- Week 4 (September 23): Sources of Data Biases and Pre-Processing Fairness Approaches

• Paper assignments:
  • The Values Encoded in Machine Learning Research
  • Flexibly Fair Representation Learning by Disentanglement
  • On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?

---

5- Week 5 (September 30): In-processing and Post-Processing Fairness Approaches and Fairness Beyond Classification

• Project meetings

---

6- Week 6 (October 7): Privacy-Preserving Machine Learning and Differential Privacy

• Paper assignments:
  • Membership Inference Attacks against Machine Learning Models
  • The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks

---

7- Week 7 (October 14): DP-SGD and PATE

• Paper assignments:
  • Overlearning Reveals Sensitive Attributes
  • Differentially Private Empirical Risk Minimizationunder the Fairness Lens

---

8- Week 8 (October 21): Reading Week

• No Class

---

9- Week 9 (October 28): Federated Learning

• Paper assignments:
  • Distributed learning of deep neural network over multiple agents
  • Comprehensive Privacy Analysis of Deep Learning: Stand-alone and Federated Learning under Passive and Active White-box Inference Attacks

---

10- Week 10 (November 4): Intorduction to Robustness and Security Attacks and Defences in Machine learning

• Paper assignments:
  • Concrete Problems in AI Safety
  • Maintaining fairness across distribution shift: do we have viable solutions for real-world applications?

---

11- Week 11 (November 11): Interpretability and Explainability

• Paper assignments:
  • Understanding Black-box Predictions via Influence Functions
  • Explaining models: an empirical study of how explanations impact fairness judgment

---

12- Week 12 (November 18): Post-hoc Explanations: Model Agnostic Methods

• Paper assignments:
  • Counterfactual Explanations for Machine Learning: A Review
  • Towards Realistic Individual Recourse and Actionable Explanations in Black-Box Decision Making Systems

---

13- Week 13 (November 26): Post-hoc Explanations: Model Specific Methods

• Paper assignments:
  • No more paper assingments!

---

14- Week 14 (December 2): Project Preperation and Neurips

• No Class

---

15- Week 15 (December 9): Project Presentation

• Room: Bleu, Salle Rona (Côte-Sainte-Catherine 1er étage, capacité 30)