STA 9890 - Course Syllabus

All syllabus and course schedule provisions subject to change with suitable advance notice.

Instructor

Professor Michael Weylandt
Department of Information Systems & Statistics
Zicklin School of Business
Baruch College, CUNY

Course Meetings

Lecture / Lab Sessions

Tuesdays 6:05pm - 9:00pm
- Baruch Main Campus (1 Bernard Baruch Way)
- Newman Vertical Campus (NVC) 5-175

Office Hours

In-Person
- Baruch Main Campus (1 Bernard Baruch Way)
- Newman Vertical Campus (NVC) 11-246
- Tuesdays 4:30-5:30pm
- Subject to periodic cancellation (communicated via Brightspace)
Virtual:
- Thursdays 4:30pm-5:30pm
- Zoom link provided via Brightspace

Grading

STA 9890 is a(n advanced) master’s-level course in Statistical Machine Learning. As such, by the end of the course, students should be able to capably apply standard machine learning techniques to important scientific and business applications, to extend and adapt standard machine learning techniques in novel fashions, to implement complex estimation and prediction algorithms, and to critically engage with modern machine learning literature. The grading in this course reflects this diversity of objectives.

Specifically, the grading in this course draws on three separate modes of assessment:

in-class ‘midterms’ which assess students’ fluency with the theoretical and mathematical aspects of machine learning;
three ‘research reports’ designed to build students’ skills in the computational and methodological aspects of machine learning; and
a course prediction competition, designed to build students’ skills in applications of machine learning to important scientific problems.

All three elements are equally weighted as follows:

33.3% One-Hour Mid-Semester Tests (Best two of three; 100 points each, 200 points total)
- Test 1: 2025-03-04 - Theory of Supervised Learning; Regression
- Test 2: 2025-04-08 - Classification; Ensemble Learning; Tree-Based Methods
- Test 3: 2025-05-06 - Unsupervised Learning
33.3% Research Reports (Best two of three; 100 points each, 200 points total)
- Report 1: Due 2025-03-07
- Report 2: Due 2025-04-18
- Report 3: Due 2025-05-09
33.3% Prediction Competition (200 points total)

Final course grades will be curved in accordance with relevant program, departmental, school, and college policies.¹

Weekly Quizzes

In lieu of homework, I will provide a set of suggested readings following each lecture. At the start of the following lecture, a short three question (\(\approx\) 15 minute) quiz will be administered. The quiz questions will not be verbatim from the reading, but if you can answer the textbook questions in the suggested sections quickly and fluently, the quiz should pose little difficulty.

These in-class weekly quizzes will generate extra credit applied to your final aggregate score. Each quiz will receive a score out of 3 added directly to your final score. Because the final aggregate score is out of 600, perfect scores on all 9 weekly quizzes can raise your final aggregate score (pre-curve) up to 5%.

To take part in the weekly quizzes, please come to class with both i) a black or blue pen; and ii) a red pen (for peer grading) each week.

Make-up opportunities for the weekly quizzes will only be allowed in exceptional and unforeseeable circumstances.

Verification of Enrollment

CUNY mandates a Verification of Enrollment (VoE) activity be completed for students to remain enrolled in a course. For STA 9890, this VoE activity takes the form of an ungraded, but mandatory, Brightspace quiz. This quiz covers the contents of this syllabus.

Students who do not successfully complete this quiz by the required date (2025-02-12) may be involuntarily disenrolled from this course.

Regrading Policy

If you feel an assignment has been improperly graded, please contact the instructor by private message on the course discussion board within 48 hours of the graded assignment being returned. Note that the instructor will regrade the assignment de novo, so your grade may be adjusted upwards or downwards.

Late Work Policy

Students are given 4 Late Days that can be used at any time during the semester, without instructor pre-approval or permission, on any of the submitted elements of this course (competition final report, research reports). When late days are used, work is accepted without penalty; late work submitted beyond the four late days will not be accepted. Late Days are used on a “first come, first serve” basis, so if 2 Late Days are used on each of the first two research reports, any late submission on the final research report will not be accepted and will receive an zero.

Additionally, note that:

Outside of the use of Late Days, no late work will be accepted. You can use Late Days however you see fit. I recommend saving them to the end of the semester for personal emergencies. Even if not needed for personal emergencies, you will appreciate having them available to you to help navigate the end-of-semester “crunch” period.
Late Days cannot be applied to in-class activities such as tests or presentations.
Late days will not be applied fractionally: 1 minute late is 1 day late; 24 hours and 1 minute late is 2 days late; etc. Brightspace will be used to determine time of submission.
Work must be submitted on Brightspace unless specifically otherwise stated in the written course description or approved in writing by the instructor at least 48 hours before the deadline. (If we discuss alterations in office hours or after class, you must follow up by email and get my pre-approval in writing.) Email submissions are not allowed.²
Technology issues are not an approved excuse for late submissions; in particular, Brightspace issues are not an approved excuse for late submission of the Research Reports. Brightspace support is provided by BCTC and the BCTC Help Desk. The Help Desk is not staffed 24 hours per day, so you are encouraged to submit early and submit often to avoid issues.

Individual exceptions will be made i) with prior written approval of instructor; or ii) ex post with written note from the Office of the Dean of Students.

Attendance Policy

Attendance is not required for this course and absences do not need to be excused. Make-up opportunities for Quizzes and Tests missed while absent will not provided except with i) with prior written approval of instructor; or ii) ex post with written note from the Office of the Dean of Students.

Pre- and Post-Reading Suggestions

Students learn material most effectively when exposed to it on multiple occasions, ideally using alternative presentations strategies and formats.³ To this end, suggested pre-reading and post-reading is provided for each week of the course. Students are encouraged to pre-read the recommended text, which typically presents that week’s material in a less technical / more intuitive manner, before each week’s course session. Similarly, students are encouraged to review the post-reading for each week after lecture to see additional examples of topics covered.

While lectures will focus primarily on ‘big picture’ and ‘major themes’, the recommended reading, especially the post-reading, provides additional coverage of relevant technical detail. Note that pre- and post-reading are entirely optional and will not be directly assessed. I have included (overlapping) presentations from various text books to encourage students to find the style and presentation that works best for them.

Tentative Course Schedule

Week	Lecture Date	Topics	Pre-Reading	Post-Reading	Research Reports	Additional Notes
1	2025-01-28	Course Overview & Introduction to ML: Taxonomy of Learning In-Sample and Out-of-Sample Accuracy Complexity and Overfitting Nearest Neighbor Methods	None	ISL §2 HR §1-4 UML §1-2, 5, 19
2	2025-02-04	Regression I: Review of OLS Linear Algebra Review Introduction to Convex Optimization	DFO §2.1-2.6, §3.1-3.4, §5.1-5.2, §5.8, §7.1, §7.3, §9.1-9.2 ISL §3	BV §2-4 HR §5, 14 UML §12, §14	Research Report #01 released on 2025-02-04 - due on 2025-03-07
3	2025-02-11	Regression II: Regularization Ridge Regression Sparsity and the Lasso Supervised Model Selection	DFO §8.1-8.3 ISL §6	BV §6 SLS §2, 4, 5 PML-1 §11.1-11.4 UML §11, 13, 25.1		VOE Syllabus Quiz due on Brightspace 2025-02-12.
	2025-02-18	*No class: Baruch on Monday Schedule (President’s Day Make-Up)*
4	2025-02-25	Regression III: Non-Linear Methods Kernel Trick Splines	DFO §12.4 ISL §7	PML-1 §11.5-11.8, §17.1 UML §16
5	2025-03-04	Mid-Term Test I: Regression Introduction to Classification
6	2025-03-11	Classification I: Basics of Classification Mixture Methods & Generative Methods	DFO §11.1-11.5 ISL §4	PML-1 §10, 12 UML §9, 24	Research Report #01 Due on 2025-03-07 (NB - Friday before class) Research Report #02 released on 2025-03-11 - due on 2025-04-18
7	2025-03-18	Classification II: Discriminative Methods Multi-Class Classification Fairness in Machine Learning	DFO §12.1-12.6 ISL §9	SLS §3 PML-1 §9 UML §15
8	2025-03-25	Ensemble Learning & Resampling Methods	ISL §5	SF §1-3, 5 UML §10
9	2025-04-01	Tree-Based Methods	ISL §8	PML-1 §18 UML §18
10	2025-04-08	Mid-Term Test II: Classification, Ensemble Learning, Tree-Based Methods Introduction to Unsupervised Learning
	2025-04-15	*No class: Baruch Spring Recess*
11	2025-04-22	Unsupervised Learning I Dimension Reduction Principal Components Analysis Covariance Estimation	DFO §3.5, §3.8, §4.2-4.6, §10.1-10.8 ISL §12.1-12.3	SLS §7-8 PML-1 §20.1-20.2 UML §23.1	Research Report #02 Due on 2025-04-18 (NB - Friday before class) Research Report #03 released on 2025-04-22 - due on 2025-05-09
12	2025-04-29	Unsupervised Learning II Clustering Density Estimation Outlier Detection Manifold Learning	ISL §12.4-12.5	PML-1 §20.3-21.6 UML §22
13	2025-05-06	Mid-Term Test III: Unsupervised Learning Introduction to Generative Models
14	2025-05-13	Course Project Presentations Course Wrap-Up				Research Report #03 Due on 2025-05-09 (NB - Friday before class)

Workload Expectations

The following approximate breakdown of expected course workload is intended to help you properly prepare for and schedule the out-of-class work associated with this course. Note that, persuant to relevant Federal and State regulations, a 3-credit course taken over a 15 week semester should require approximately 6 hours of out-of-class work from an average student, or 90 hours total over the course of the semester.⁴ ⁵ This time should be roughly allocated as:

Weekly Pre-Reading (9 hours - 1 hour per lecture)
Weekly Post-Reading and Review (18 hours - 2 hours per lecture)
Test Preparation (15 hours total - 5 hours per test)
Research Reports (27 hours total - 9 hours each)
Prediction Competition (21 hours total)
- Ongoing Submissions (18 hours - approximately 2 hours per week for 9 weeks)
- Final Presentation and Report Preparation (3 hour)

Note that, for the course prediction competition, you will need to make regular progress throughout the semester. If you attempt to “back-load” your work, you will do poorly.

Coding Requirements

STA 9890 is, at its heart, a machine learning course and as such use of a machine, i.e. coding, is required, even though there is no formal coding prerequisite for this course. Per the External Resources Policy, you are allowed (and encouraged) to use freely available coding assistance technologies, including generative tools like GitHub CoPilot.⁶

You may use whatever programming language you prefer to complete the course assignments, subject to instructor approval: R, python, julia, and matlab are pre-approved for all students by default. (I will approve most other languages as well, provided they are not too obscure.)

Tools like quarto or Jupyter Notebooks will be useful for completing the Research Reports required for this course. If you have not used these previously, many useful free resources can be found online, including my STA9750 course materials.

`PyTorch`, `TensorFlow`, and Similar

An exciting aspect of modern ML is the widespread availability of flexible and powerful model fitting frameworks like TensorFlow and PyTorch. Use of these frameworks is not required for this course, though students interested in ML focused careers would benefit from familiarizing themselves with them.

For the Research Reports, you may not use these frameworks in lieu of the required coding aspects of each assignment. For example, if a Report requires implementing a gradient descent method, you must do so “by hand:” computing gradients analytically, selecting step sizes, performing updates, checking convergence, etc.⁷ In Reports where an optimization library like CVX is used, you may use these frameworks instead, though doing so will require a bit more work. Each Report allows (and encourages!) you to go beyond the standard instructor-provided skeleton; comparing your implementation with an implementation using one of these frameworks would be an interesting contribution, so you are welcome to do so in addition to the “by hand” implementation.

For the Prediction Competition, you may use these frameworks or indeed any freely available (F/OSS) statistical or ML software; instructor pre-approval is not required.