CMPT 727: Statistical Machine Learning. Spring 2020.Course logistics
- Tuesdays 11:30AM - 12:20PM, AQ3153
- Thursdays 9:30AM - 11:20AM, AQ3159
Instructor: Maxwell Libbrecht
- Web page
- Office: TASC 1 9219
- Email: maxwl at sfu dot ca
- Office hours: Tuesdays 9:45-10:30am. Blackboard link
- Heng Liu (liuhengl at SFU dot ca)
- Office hours: 12 pm to 1 pm, every Wednesday.
- Office hour location: ASB9810
Textbook website: Machine Learning: a Probabilistic Perspective by Kevin Patrick Murphy.
- W2 Jan 6-12:
- Reading: Chapter 1.
- W3 Jan 13-19:
- Reading: 2.1, 2.2, 2.3, 2.4-2.4.1 (optional: 2.4.2-7), 2.5-2.5.2 (optional: 2.5.3-4),
- Thu: Quiz 1
- W4 Jan 20-26:
- Reading: 3.1-3.3, 3.5.
- W5 Jan 27-Feb 2:
- Reading: 5.1-5.3.1 (optional: 188.8.131.52, 184.108.40.206), 5.7.1-220.127.116.11, 6.1, 6.5 (optional: 6.5.4-5).
- Thu: Quiz 2
- W6 Feb 3-9:
- Reading: 10.1-10.4 (optional: 10.2). 11.1-11.2.3. 11.3-11.3.1. 11.4-11.4.4 (optional: 18.104.22.168, 11.4.3). 11.5-6.
- W7 Feb 10-6:
- Reading: 10.5. 19.1-2, 19.4-19.4.4, 20.1-2 (optional: 20.2.3-4), 20.5.
- Thu: Quiz 3
- W8 Feb 17-23: No lecture (reading break).
- W9 Feb 24-Mar 1:
- Reading: 23.1-23.5.6, 24.1-24.2 (optional: 24.2.4, 24.2.7).
- Thu: A1 due
- W10 Mar 2-8:
- Reading: 24.3 (optional: 24.3.6-7). 24.4 (optional: 24.4.2). 24.6-24.6.1. 21.1-3 (optional: 21.2.2).
- Thu: Quiz 4
- W11 Mar 9-15:
- Reading: 21.5 (optional: 22.214.171.124), 21.7, 22.1-22.2.5.
- W12 Mar 16-22:
- Lecture links: Tuesday (recording). Thursday (recording).
- Reading: 25.1-3.
- Fri: A2 due
- W13 Mar 23-29:
- W14 Mar 30-Apr 3:
- W15 Apr 6-11:
- Tue: Final exam. 1-4pm (remote).
- Lectures will be remote starting W12-Tue (March 17). We will use Blackboard Ultra through Canvas.
- See the schedule above for lecture and recording links. Alternatively, you can join the course on canvas.sfu.ca in the "Bb Collaborate Ultra" tab to join the lectures.
- I will do my best to minimize the impact of the switch to remote instruction. I know that not everyone has access the resources they need to participate fully -- fast internet connection, quiet space at home, etc -- and may be handling other issues related to the covid-19 outbreak. I will aim to make assessment criteria as fair as possible given the circumstances. Please do the best you can to engage with the course; however, you should prioritize your health and well-being.
- Chat will be enabled during the lecture. You can send messages to the whole group or directly to me. I will watch chat during lecture and answer questions as usual. (You do not need to use the "raise hand" feature.)
- We will have in-class exercises as usual. Blackboard has a feature for breakout groups, which we will use enable our usual discussion in small groups. Please chat with your breakout group about your ideas for the exercise. I will enter the chat from each breakout group in turn and ask one person to summarize the thoughts of the group. The remaining participation grade will be determined by participation by chat in the lecture.
- Office hours will work the same as lecture, except that participants will be able to write on the virtual whiteboard and share their screen and audio.
- Remaining exams will be remote. They will be timed and paper-notes-only as usual. There will be 15 minutes allocated at the end for scanning and uploading your exam it. Please respect your classmates by following these rules. I know that some people will use this as an opportunity to cheat; I will take this into consideration when determining the curve.
Lecture format: Lectures will emphasize interactive problem solving in class. 15-20 minutes per class hour will be lecture format to introduce the material in the next chapter. The rest of class time will be spent solving questions in groups. Your groups will be based on assigned seating: see here (coming soon). Participation is mandatory.
Reading: Textbook reading assignments will be posted in the schedule. Reading will be necessary to learn the course content, as lecture will be spent applying knowledge gained from the reading. Reading is due on Thu of the corresponding week. That is, if Chapter 3 is the reading for Week 4, then lectures in W3-Thu and W4-Tue will introduce Chapter 3, and in-class questions on W4-Thu and W5-Tue (or a quiz on Q4-Thu) will ask about Chapter 3 content.
Quizzes: Every other Thu, for the first 40 mins of lecture. There will be an optional homework associated with each quiz. Any paper notes are allowed; no laptop, phone or calculator.
Coding assignments: There will be three coding assignments. Late policy: You get three grace days. Turning in a coding assignment late costs one grace day per day after the due date (rounded up). If you run out of late days, late submission costs 10% of the assignment grade per day. No submissions will be accepted more than five days after the due date.
How you should spend your time in this course (12 hr/week total):
- 3 hr/week: Lecture attendance.
- 3 hr/week: Reading.
- 4 hr per quiz (avg 2 hr/week): Optional homework and quiz preparation.
- 10 hr per assignment (avg 2.5 hr/week): Coding assignments.
- 13 hrs (avg 1 hr/week): Review and final exam preparation.
Unofficial prerequisites: No official prerequisites. However, the course assumes knowledge of machine learning (CMPT 726) and linear algebra (MATH 240). The course is open to advanced undergradutes with permission.
How to enroll as an undergraduate: Please follow the following steps: (1) Come to the first lecture, where I will explain the expected background. (2) Email me stating that you have taken a course in probability/statistics and a course in machine learning (CMPT 726). If it is appropriate, I will respond with my approval. (3) Fill out the prerequisite waiver form and attach a screenshot of our email exchange.
Auditor policy: Anyone is welcome to sit in on lectures, provided there are sufficient seats.
Other good resources.
- Introduction. Machine learning: what and why? Supervised learning. Unsupervised learning. Basic concepts in ML.
- Probability. Review of probability theory. Common discrete and continuous distributions. Joint probability distributions. Transformations of distributions and the central limit theorem.
- Bayesian and frequentist statistics. Bayesian: prior, posterior, likelihood. Frequentist: maximum likelihood estimate (MLE).
- Bayesian networks. Inference, learning parameters. Conditional independence properties. Exact inference.
- Mixture models and the EM algorithm.
- Sampling and sampling-based inference. Rejection sampling, Gibbs sampling, Markov Chain Monte Carlo and Metropolis Hastings.
- Variational inference. Neural networks and variational autoencoders.
- 45% Coding assignments (15% each).
- 35% Biweekly quizzes (7% each). Your lowest grade (after the curve) will be dropped when calculating your grade.
- 15% Final exam.
- 5% Participation in class.
- +5% Extra credit for discussion on Piazza. Max 20 points. 1 point = first answer to a simple question (logistics etc). 3 points = Accurate, in-depth response to a content question or in-depth discussion of AI-related topic. 3 points = First to point out nontrivial error in course materials (assignments, lecture slides, etc). 0 points = Asking a question to help yourself with assignments etc; "meta" discussion (logistics etc); response that is incorrect or unintelligible.