CPSC 440/550: Advanced Machine Learning – 23W2 (Jan-Apr 2024)

Schedule

• Calculus
• Max and Argmax
• Probability
• Linear Algebra
• Norms
• Gradient and Hessian of Quadratics
• Positive Semi-Definite Matrices
• Gaussians with Conjugate Priors
• Expectation Maximization
• Forward-Backward for HMMs

Italicized entries are tentative; in particular, the timing and even number of assignments might change. Textbook acronyms are explained below.

Overview

This course is intended as a second or third university-level course on machine learning, a field that focuses on using automated data analysis for tasks like pattern recognition and prediction. The class is intended as a continuation of CPSC 340 (also called 540, or previously 532M); it will assume a strong background in math and computer science. Topics will (roughly) include deep learning, generative models, latent-variable models, Markov models, probabilistic graphical models, and Bayesian methods.

Note that the numbers for graduate cross-listings of our machine learning courses changed this year: previously 340 was also called 532M, and 440 was also called 540. Now 340 is also called 540, and 440 is also called 550.

Logistics

The course meets in person in MacMillan 360. I plan to release recordings, but encourage you to come to class in person if you can.

Grading scheme:

• 40% assignments (a mix of conceptual questions, math, and programming). A1 must be done alone; later assignments can be done alone or with a partner.
• 20% regular quizzes (short conceptual and definitional questions)
  • Self-book a time slot in the computer-based testing facility; instructions on Piazza.
  • One quiz can be skipped: just don't book a slot, in which case your grade is the average of 5 quizzes instead of 6. Note this is not the same as “drop the lowest score”; if you take the quiz, it counts.
• 40% from final assesements, which will be one or both of:
  • Exam: a traditional 2.5 hour final exam, handwritten, in person.
  • Project: a course project, relatively limited scope, done in a group of 2-3 (details announced later).
  440 students must do at least one of these; if they do both, they'll receieve the maximum of the two grades. 550 students must do both, and will receieve the mean of the two grades.

Further details in the syllabus slides.

Registration and Prerequisites

• The majority of the seats in 440 are reserved for UBC computer science majors. For other students, to enroll in the course you need to sign up for the wait list.
• The seats in 550 are reserved for graduate students in CPSC, EECE, or STAT. If these seats are full or you are from a different department, to enroll in the course you need to sign up for the wait list. Please read about the differences between 340/540 and 440/550 below, and do not take this course "because 340/540 is full" or not offered in a given term.

Starting in the second week of classes, we'll have weekly tutorials run by the TAs. These will do things like go through provided assignment code, review background material, review big concepts, and/or do exercises. You can register for particular tutorial sections if you want to save a seat at a particular time, but note that you do not need to register in a tutorial section.

CPSC 340/540 vs. CPSC 440/550: CPSC 340 and 440 are roughly structured as one full-year course. CPSC 340 (which is sometimes cross-listed as CPSC 540 for graduate students; formerly 532M) covers more data mining methods and the methods that are most widely-used in applications of machine learning. CPSC 440 (cross-listed as CPSC 545 for graduate students) focuses on probabilistic methods which appear in more niche applications, as well as various other topics not covered in 340/540. It is strongly recommended that you take CPSC 340/540 first, as it covers the most fundamental ideas as well as the most common and practically-useful techniques. In 440/550 it will be assumed that you are basically familiar with all the material in the current offering of CPSC 340/540. Note that online machine learning courses and courses from many other universities may not be an adequate replacement for CPSC 340; they typically have more overlap with our applied machine learning course, CPSC 330. If you're not sure, look at last term's 340 website and see if it all seems familiar.

Resources

Textbook: There is no textbook for the course, but the textbook with the most extensive coverage of many of the course's topics is Kevin Murphy's Probabilistic Machine Learning series. While the one-volume 2012 version covers most of the material, we'll refer to the very recent two-volume version (2022/2023), PML1 and PML2, both of which have free Creative Commons draft pdfs through those links. I'll try to refer to the relevant sections of both versions as we go, as well as links to various other free online resources.

If you need to refresh your linear algebra or other areas of math, check out Mathematics for Machine Learning (Marc Deisenroth, Aldo Faisal, Cheng Soon Ong; 2020).

Related courses: Besides CPSC340, there are several 500-level graduate courses in CPSC and STAT that are relevant: check out the graduate courses taught by people on the ML@UBC page and the MILD list. CPSC 422/425/436N, DSCI 430, EECE 360/592, EOSC 510/550, and STAT 305/306/406/460/461 are also all relevant.

Some related courses that have online notes are:

• Machine Learning (UBC 2013)
• Introduction to Machine Learning (Alberta)
• Machine Learning (MIT)
• Course in Machine Learning (Maryland)
• Apprentissage (ENS, in French)

• Machine Learning (Mathematical Monk)