Machine Learning for Personalization

This course will cover advanced machine learning methods that can be used to personalize the user experience in order to achieve better conversion, engagement, retention, and other metrics. Topics include collaborative filtering, ranking, page construction, search, targeted advertising, presentation/production/popularity bias, and more. The techniques that will be leveraged include matrix factorization, probabilistic modeling, Bayesian modeling, graph-based learning, semi-supervised learning, explore/exploit online learning, inverse propensity de-biasing, incremental modeling, and causal learning. Students are expected to have taken Machine Learning COMS 4771 or equivalent course.

Note for CS students: This course and its credits (e.g. towards the degree) will be considered equivalent to COMS 4772 Advanced Machine Learning.


Week 1: Overview and Matrix Factorization for Collaborative Filtering.
Matrix Completion has No Spurious Local Minimum

Week 2: Probabilistic Matrix Factorization.
Algorithms for Non-negative Matrix Factorization
Projected Gradient Methods for Non-negative Matrix Factorization
Topic Models
A Collapsed Variational Bayesian Inference Algorithm for Latent Dirichlet Allocation

Week 3: Nonlinear Probabilistic Embedding.
AutoEncoding Variational Bayes
Variational Autoencoders for Collaborative Filtering

Week 4: Optimizing Rank Loss, Diversity, Page Construction.
Ranking via Robust Binary Classification
Adaptive, Personalized Diversity for Visual Discovery
Learning a Personalized Homepage

Week 5: Multi-Armed Bandits for Explore/Exploit.
Practical Evaluation and Optimization of Contextual Bandit Algorithms

Week 6: Contextual Bandits for Explore/Exploit continued.
A Contextual-Bandit Approach to Personalized News Article Recommendation
An Empirical Evaluation of Thompson Sampling
Artwork Personalization at Netflix

Week 7: Offline Policy Evaluation, Replay, IPS.
Unbiased Offline Evaluation of Contextual-bandit-based News Article Recommendation Algorithms
Doubly Robust Policy Evaluation and Learning
Practical Evaluation and Optimization of Contextual Bandit Algorithms

Week 8: Offline Policy Evaluation, Replay, IPS Part 2.
The Self-Normalized Estimator for Counterfactual Learning

Week 9: Causality, Incrementality and Two-Stage Least Squares.
Instrumental Variables and Two-Stage Least Squares
Instrumental Variables, 2SLS and GMM

Week 10: Instrumental Variables in Large, Deep and Online Models.
Deep IV: A Flexible Approach for Counterfactual Prediction.

Week 11: IV Bandits and Social Network-Based Machine Learning.
Instrument-Armed Bandits.
Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions.
Local and Global Consistency.
Graph Construction and b-Matching for Semi-Supervised Learning.

Week 12: Project Presentations Part 1

Week 13: Project Presentations Part 2

Week 14: Project Writeups Due

Homework 1. Time frame: 3 weeks.
Homework 2. Time frame: 3 weeks.
Project. Time frame: 8 weeks.

40% of the grade will be based on two homeworks. The remaining 60% will be based on a conference-style novel paper and presentation. Teams of 3 people will be assembled for a class project (if you are less than 3 people, you will still be evaluated the same way).