Rensselaer Polytechnic Institute
Department of Computer Science


CSCI 6100/4100: Machine Learning

Fall 2007

OVERVIEW

This course is an introduction to many of the important themes in machine learning, including supervised and unsupervised learning, online learning, reinforcement learning, and learning in societies of agents. It will focus on statistical methods, how to evaluate success, and learning as a critical component of decision-making.

Instructor: Sanmay Das
Office: Lally 302
Phone: x2782
Office hours: Tuesdays from 1:30-2:30 PM and Wednesdays from 3:00-4:00 PM, or by appointment.

SYLLABUS

The syllabus that was handed out on the first day of class is available here.

LECTURES

Lectures will be on Tuesdays and Fridays from 12pm to 1:20pm in Low 3130.   
Date Topics Notes
Aug 28 Introduction to supervised, unsupervised, reinforcement, and "rational" (Bayesian) learning. Review of probability, statistics, and linear algebra. PDF notes
Aug 31 Estimators. Maximum-likelihood and Bayesian estimation PDF notes
Sep 4 The bias/variance tradeoff and linear models for regression PDF notes
Sep 7 Generative and discriminative models: Naive Bayes and logistic regression Notes are from Tom Mitchell's draft chapter, which is available here
Sep 11 Logistic regression, continued. Overfitting. Evaluating classifiers. Precision, recall, and ROC curves. PDF notes
Sep 14 Basics of utility theory. Cost-sensitive classification. PDF notes
Sep 18 Separating hyperplanes. The perceptron algorithm and the perceptron convergence theorem. Intro to the SVM. PDF notes
Sep 21 Support vector machines. The "kernel trick." Regularization. Notes are included in the previous lecture's.
Sep 25 Decision tree induction and pruning PDF notes
Sep 28 Nearest neighbor methods. Ensemble methods: boosting and bagging PDF notes
Oct 2 Papers: Domingos (D) and Bousquet & Elisseeff (BE) D is here (RPI access) and BE is here
Oct 5 Expectation-Maximization and k-means PDF notes
Oct 9 No class (Monday schedule)
Oct 12 Midterm
Oct 16 Markov Models: Filtering, Smoothing, and Most Likely Path PDF notes (incl. Kalman filters)
Oct 19 Kalman filters. Markov Decision Processes PDF notes (incl. next lecture)
Oct 23 MDPs continued. The Bellman Equation. Dynamic programming Check previous lecture for notes!
Oct 26 No class: Attend CS Day
Oct 30 Bandit problems PDF notes
Nov 2 Reinforcement learning PDF notes
Nov 6 No class (SD at INFORMS)
Nov 9 Secretary problems Notes are from Chapter 2 of Thomas Ferguson's book, available here
Nov 13 More optimal stopping. Stable marriage. Stable marriage PDF notes
Nov 16 Papers: Das and Kamenica (DK) and Vermorel and Mohri (VM) DK is available here and VM is here
Nov 20 Learning in sequential search Notes are from this paper
Nov 27 Sequential search contd. Fire alarm.
Nov 30 Online learning and the experts framework. Intro to game theory. Online learning and experts notes
Dec 4 Game theory contd., and some learning in games. Game theory notes
Dec 7 Second exam
 


TEXTBOOKS

The course will not be based on any single book.   The following textbooks are all recommended as basic references:

Many of the topics we cover may differ significantly in coverage from any of the texts mentioned above. Therefore it is important to come to lectures and take notes.

ASSIGNMENTS

Assignment 1: Due Sep 21
Assignment 2: Due Nov 9
Assignment 3: Due Dec 7