snisarg/ml-class-iap2017

A Whirlwind Tour of ML

###IAP 2017 course at MIT

About

This course gives a high-level overview of diverse areas of machine learning. The goal is to introduce students to core concepts and techniques in ML, and provide enough of a primer on different sub-areas of ML so that students can choose the right approach for a given problem and explore interesting topics further.

The course covers an introduction to ML, Inference, Bayesian Methods and Neural Networks. Each class is taught by graduate students or post-docs at MIT working in the specific areas.

Organized by Manasi Vartak and Maggie Makar from MIT CSAIL.

Session I: Introduction to ML

This session gives an overview of supervised and unsupervised learning, and an introduction to probabilistic graphical models.

Concepts: Loss functions, Linear regression, Logistic regression, SVMs, Decision trees, Random Forests, Clustering, PCA, Graphical Models, Variable Elimination

Taught by Manasi Vartak.

Slides

Resources

• MIT 6.867 Machine Learning
• Coursera Machine Learning
• MIT 9.520 Statistical Learning Theory
• CMU: Intro to Machine Learning
• Michael Jordan Review of Graphical Models
• Coursera Probabilistic Graphical Models
• Columbia University: Probabilistic Graphical Models

Session II: Inference

This session gives an overview of (approximate) inference for probabilistic graphical models.

Concepts: Gaussian Mixture Models, Variational Inference, Monte Carlo Sampling

Taught by Maggie Makar ###Slides

Resources

• Tutorial on VI
• A Review of recent work on VI, Section 5
• Tutorial on Sampling methods
• A review (and really cool demos) of recent work on sampling

Session III: Bayesian Methods

This session gives a whirlwind tour of Bayesian Methods in ML.

Concepts: What does it mean to be Bayesian in ML, Why be Bayesian, Posterior Inference, Parameteric vs. Non-Parametric Bayes

Taught by Trevor Campbell ###Slides

Resources

See slides!

Session IV: Neural Networks

This session gives an overview of neural networks, particularly as applied to computer vision.

Concepts: Neural Nets, Convolutional NNs, AlexNet, GoogleLeNet, Transfer learning

Taught by Carl Vondrick ###Slides Note: these slides are not exactly the ones that were presented in class. Please feel free to reach out to Carl if you need information that's not in these slides.