CSCI 4750/5750: Machine Learning
Saint Louis University · Spring 2027
| Meetings | Mon/Wed/Fri 10:00 – 10:50 AM · room to be announced |
| Term | Jan 11 – May 3 (last class meeting) |
| Project presentations | Fri Apr 30 and Mon May 3 (in class) and Wed May 5, 12:00–1:50 PM (final-exam slot) — no written final |
| Instructor | Jin Zhao · jin.zhao@slu.edu · Ritter Hall 309 |
| Office hours | Mon/Wed/Fri 11:00 AM – 12:00 PM in Ritter Hall 309, and by appointment |
| Listings | CSCI 4750-01 (undergraduate, “Machine Learning”, CRN 23042) · CSCI 5750-01 (graduate, “Introduction to Machine Learning”, CRN 23043) |
| In-class questions | jinzhao3611.github.io/csci4750/q — open this in class (or scan the QR code shown on screen) |
Course Description
This course introduces the field of machine learning, with emphasis on the probabilistic models that dominate contemporary applications: how computers learn from examples, classification and regression, and supervised and unsupervised learning, together with the statistical models and computational methods behind them.
The course is about fundamentals: every method is derived, worked by hand on small numbers, and implemented from scratch before you use a library version. It is also about now: each lecture closes by showing where its classical idea lives in 2026–27 industry and research — the softmax regression inside a language model, the nearest-neighbor search behind recommendations, the gradient boosting inside fraud detection, the diffusion models behind image generators, the reinforcement learning behind reasoning models. The arc runs from k-nearest neighbors in week one to foundation models in April, measured all semester on one shared benchmark, the Fashion-MNIST scoreboard.
The course is lecture-oriented, with four substantial programming assignments, an in-class midterm, and a final project (teams of 1–4) presented on the three presentation days. Prerequisites: STAT 3850, CSCI 2100, and MATH 2530. Linear algebra beyond the basics is not assumed — Lectures 3–4 teach the math toolkit the course uses.
See the class schedule for the lecture-by-lecture calendar with readings and due dates.
Learning Goals
- Formulate a prediction problem as learning from data — features, labels, hypothesis class, loss, and evaluation — and explain the difference between fitting the training data and generalizing.
- Derive and implement core supervised methods from scratch — nearest neighbors, linear and logistic regression, regularized models, Naive Bayes, decision trees — and explain how kernels, SVMs, and ensembles extend them.
- Connect losses to likelihoods: explain maximum likelihood, MAP estimation, and Bayesian inference, and use the probabilistic view to choose and justify a model.
- Apply unsupervised methods — k-means, Gaussian mixtures with EM, PCA, embeddings, matrix factorization — and interpret their output honestly.
- Build, train, and debug neural networks, from backpropagation by hand to convolutional networks in PyTorch, and explain how self-supervised learning, generative models, and reinforcement learning power today's foundation models.
- Evaluate models rigorously — cross-validation, appropriate metrics, leakage checks, fairness criteria, and distribution shift — and recognize when a reported number should not be trusted.
- Conduct and communicate an original ML project: formulate a question, run experiments against baselines, and present findings in a written report and a talk.
Grading
| Component | Weight | Notes |
|---|---|---|
| Homework (4 × 10%) | 40% | From-scratch NumPy implementations + library comparison + short written analysis (graduate extensions required in 5750) |
| Midterm exam | 20% | In class Wed Mar 3; closed book — every problem comes from a set released one week ahead; covers Lectures 1–19 |
| Final project | 30% | Teams of 1–4; proposal 5%, presentation 10% (Fri Apr 30, Mon May 3, or the Wed May 5 exam slot), report 15% |
| Participation (in-class questions) | 10% | Live questions answered in class on your own device (QR code shown in class); credit for participating, not correctness; miss up to 6 sessions without penalty |
Homework specifications are on the assignments page; project tracks, timeline, and rubrics are on the project page. Full grading details, late policy, AI-tools policy, and university statements are in the syllabi: CSCI 4750 · CSCI 5750.
Materials
All required materials are free.
- [ISLP] James, Witten, Hastie, Tibshirani & Taylor, An Introduction to Statistical Learning with Applications in Python (2023) — statlearning.com
- [PML] Murphy, Probabilistic Machine Learning: An Introduction (2022) — probml.github.io/pml-book
- [MML] Deisenroth, Faisal & Ong, Mathematics for Machine Learning (2020) — mml-book.github.io
- [UDL] Prince, Understanding Deep Learning (2023) — udlbook.github.io/udlbook
- [S&B] Sutton & Barto, Reinforcement Learning: An Introduction, 2nd ed. (2018) — incompleteideas.net
- Papers and posts for the “Now” segments and the modern-ML lectures — listed on the class schedule.
Assignments are sized for Google Colab. A GPU helps only in HW4 and in some projects, and Colab provides one.
Communication
Lecture slides, homework, and announcements are posted on Canvas. In-class questions run at jinzhao3611.github.io/csci4750/q — a QR code is shown in class, but you can bookmark that page too. For questions or concerns related to the course, email jin.zhao@slu.edu or come to office hours (Mon/Wed/Fri 11:00 AM – 12:00 PM, Ritter Hall 309).