COMS 4771 Machine Learning

This page is dedicated to information and materials for the Columbia University COMS 4771 Machine Learning course taught by James McInerney in Fall 2017.

Lectures

Mon 2:40-3:55pm and Wed 2:40-3:55pm at 428 Pupin Laboratories.

Office Hours

Office hours will be held in 7LW1A on 7th floor CEPSR Mon 4-5pm and Wed 4-5pm from Wednesday 13th September onwards.

Office hourse dial in: As an experimental feature, please email me the day before office hours with the subject line "schedule office hours" and I will make sure a Google hangout video chat will be open at that time. The handle to call is james.office.hours@gmail.com and it will only be available during office hours. We may remove this option if the trial is not successful.

Instructional Assistants

Oscar Chang (oscar.chang) Office hours: Mondays 11:30am-12:30pm

Wei Dai (wd2281), office hours: Thursdays 9:30-11:30am

Ishan Jain (ikj2102), office hours: Tuesdays 11:00am-1:00pm

Akshay Khatri (ajk2237), office hours: Tuesdays 10:30am-12:30pm

Boqiao Lai (bl2633), office hours: Wednesdays 4-5pm

Anuj Sharma (as4529), office hours: Fridays 1-3pm

All TA office hours are held in the TA room on 1st floor Mudd Buildling. Add at-dot-columbia-dot-edu to reach them by email.

Syllabus

Non-parametric methods: nearest neighbors, K-D and decision trees

Parametric methods: generative models, linear classifiers, features and kernels, support vector machines, convex optimization, neural networks

Representation learning: dimensionality reduction, collaborative filtering, clustering, latent variable models, graphical models

Lecture Slides

Slides will be made available around the time of each lecture. Thanks to Professors Daniel Hsu and Nakul Verma, and Francisco J. R. Ruiz for allowing me to adapt their slides.

Overview, Wed Sept 6th 2017.

Nearest neighbors, Mon Sept 11th 2017.

Decision Trees, Wed Sept 13th 2017.

Maximum likelihood estimation and naive Bayes, Mon Sept 18th 2017.

Gradient descent, Wed Sept 20th 2017.

Perceptrons and kernels, Mon Sept 25th 2017.

Support vector machines, Wed Sept 27th 2017.

Parametric and non-parametric regression, Wed Oct 4th 2017.

Unsupervised learning and clustering, Mon Oct 9th 2017.

Overview of neural networks, Mon Oct 23rd 2017.

Nakul Verma's graphical models, Mon Nov 13th 2017.

Homework

Homework problem sets the same for both COMS 4771 courses this semester (the other one is taught by Nakul Verma on Tuesdays and Thursdays).

All assignments will be graded on Gradescope. Please use Entry Code MX3XV8 to add yourself to the course on there.

In contrast to the other section (and regardless of what it says on the homework) please submit your solutions as one file to GradeScope with short code inline and longer code as an appendix. To make it easier to mark, thank you for pointing towards the code using the GradeScope mechanism upon submission.

HW0 (data), due 11:59pm Sept 10th on Gradescope.

HW1 (data), due 11:59pm Oct 8th on Gradescope.

HW2 (data), due 11:59pm Oct 27nd on Gradescope.

HW3, due 11:59pm Nov 12th on Gradescope.

HW4, due 11:59pm Dec 3rd on Gradescope.

Key Dates

Exam 1: Wednesday October 18th, 2017

Exam 2: Monday December 11th, 2017

Related Sites

Please see Piazza and Courseworks.

Course Policies

Please see the end of the Overview lecture for course policies.