safr-ai-lab / resources

Onboarding Logistics

• Expectations for joining and application process here.
• Communication: via the Lab Slack, weekly 1:1 meeting, biweekly lab meetings (Thursdays at 11am).
• Project Tracking: Notion page created for each project, README on github.
• Code: All code is in project specific github repositories
• Compute: FASRC, submit a request to join our lab which is under sneel.
• Space: SEC 6th Floor, suite to the left of the elevators. Our Lab Manager Amanda Chieco ([e-mail]) will set you up with key card access and a desk.
• Seth's contact info: Morgan Hall 485, Cell: 4016327495, Email: [e-mail].

Seminars and Groups to sign up for when joining the lab. Please email the relevant person and cc me when you are signing up.

• Kempner Institute Seminar
• Boston Area DP Seminar
• Privacy Tools mailing list and weekly meeting - [e-mail]
• OpenDP mailing list and slack - [e-mail]
• Theory of Computation email list and seminar - [e-mail]
• Harvard ML Theory Seminar & Slack

Reading List

general ML topics

• Contrastive Learning from Lillian Weng
• NTK Theory from Andrea Montanari
• Log-Concave Sampling
• Pattern Recognition and Machine Learning by Christopher Bishop - standard text on ML, light on deep learning
• Probability: Theory and Examples (Durrett) - we will make frequent use of tools like the Chernoff bound or Azuma's inequality.
• Understanding Machine Learning (textbook)

online learning + computational learning theory

• An Introduction to Computational Learning Theory - standard text on PAC-learning & VC-theory
• Online Algorithms in Machine Learning
• UCB Algorithm - stochastic bandits
• Improved Algorithms for Linear Stochastic Bandits
• Tim Roughgarden notes on no-regret algorithms

special topics of interest to the lab

membership inference

• Genetic Membership Inference Attack (Homer)
• Membership Inference Attacks on Machine Learning: A Survey
• Creating Artificial Genomes using Neural Networks
• Label only membership inference attacks

differential privacy

• The Algorithmic Foundations of Differential Privacy - at least ch. 1-3, ideally ch. 4
• Gautham Kamath's lecture notes on differential privacy with videos
• Iterative Methods for Private Synthetic Data: Unifying Framework and New Methods

Machine Unlearning

• Machine Unlearning

interpretability

• Counterfactual Explanations for Machine Learning: A Review
• Transparency: Motivations and Challenges
• Explainable ML Tutorial

selected papers:

• On the privacy risks of model explanations

fairness

• Fairness and Machine Learning Book - ch. 1-3

selected papers:

• Preventing Fairness Gerrymandering
• Calibration for the Computationally Identifiable Masses