DSAA3073
Fall 2026 · HKUST-GZ
Theories in Data Science
DSAA3073 introduces the theoretical foundations of data science, covering mathematical preliminaries, PAC learning, classical methods, ensemble learning, and online learning. The course emphasizes rigorous reasoning, mathematical guarantees, and connections between theory and modern data-science problems.
Course Introduction Slides (PDF)
Course Information
| Type | Days & Times | Room | Meeting Dates | Format |
|---|---|---|---|---|
| Lecture | Mon 15:00–17:50 |
Room 202, E3 | Sep 1 – Dec 7, 2026 | Self-paced, assisted learning |
| Tutorial | Tue 09:00–09:50 |
Room 202, E3 | Sep 1 – Dec 7, 2026 | Group-based learning and support, led by TA |
Teaching Team
| Role | Name | Contact |
|---|---|---|
| Instructor | Prof. Wei WANG | weiwcs@ust.hk |
| Teaching Assistant | Xin LI | xli420@connect.hkust-gz.edu.cn |
Bring a laptop or tablet and use the provided AI app during class learning activities. Follow HKUST(GZ)'s Academic Honor Code. Academic misconduct is not tolerated.
Schedule
Lectures 1–11 cover five Parts. Lectures 12–13 are reserved for student presentations. Each content lecture has a Learning Sheet, Validation Sheet, and Cheatsheet. In the Validation column, Questions links to the question-only sheet and Solutions links to the sheet with worked answers. Items that have not yet been released are shown as TBD.
Part Test format
Each Part Test is worth 120 marks, lasts 30 minutes, and contains 4 multiple-choice questions and 6 short-answer questions. Tests are held in the class following completion of the corresponding Part.
| Lecture | Lecture Date / Test | Part | Topics | Learning Sheet | Validation | Cheatsheet |
|---|---|---|---|---|---|---|
| Lecture 1 | Sep 7 | Part 1 | Probability, concentration, norm geometry, compact SVD, Hilbert spaces | 📝 Questions 📖 Solutions |
||
| Lecture 2 | Sep 14 | Part 1 | Convexity, optimality certificates, smooth descent, Big-O and feasibility | 📝 Questions 📖 Solutions |
||
| Lecture 3 | Sep 21 🧪 Part 1 Test |
Part 2 | Risk, ERM, PAC learning, finite classes, VC dimension | 📝 Questions 📖 Solutions |
||
| Lecture 4 | Sep 28 | Part 2 | Risk, ERM, PAC learning, finite classes, VC dimension | 📝 Questions 📖 Solutions |
||
| — | Oct 5 | — | No lecture — National Day holiday | — | — | — |
| Lecture 5 | Oct 12 | Part 2 | Agnostic PAC, uniform convergence, Rademacher complexity | 📝 Questions 📖 Solutions |
||
| Lecture 6 | Oct 19 🧪 Part 2 Test |
Part 3 | Bias–variance, ill-posed problems, Tikhonov, ridge | 📝 Questions | ||
| Lecture 7 | Oct 26 | Part 3 | Cross-validation, LOOCV, PRESS, AIC, BIC | 📝 Questions | ||
| Lecture 8 | Nov 2 🧪 Part 3 Test |
Part 4 | Weak learning, AdaBoost, training-error guarantees | 📝 Questions | ||
| Lecture 9 | Nov 9 | Part 4 | Margins, exponential loss, coordinate descent | 📝 Questions | ||
| Lecture 10 | Nov 16 🧪 Part 4 Test |
Part 5 | Expert advice, Weighted Majority, MWU | 📝 Questions | ||
| Lecture 11 | Nov 23 | Part 5 | OCO, OGD, FTL, FTRL, online-to-batch | 📝 Questions | ||
| Lecture 12 | Nov 30 🧪 Part 5 Test |
Presentations | Student Presentations I — topic TBD | Presentation brief: TBD | TBD | — |
| Lecture 13 | Dec 7 | Presentations | Student Presentations II — topic TBD | Presentation brief: TBD | TBD | — |
Assessment
| Component | Weight |
|---|---|
| Five Part Tests | 70% |
| Student Presentation | 30% |
| Total | 100% |
Each Part Test is worth 120 marks, lasts 30 minutes, and contains 4 multiple-choice questions and 6 short-answer questions. The five Tests are equally weighted at 14% each.
Weekly Validation Sheets are provided for checking understanding and do not contribute to the course grade unless announced otherwise. Student presentations take place in Lectures 12–13 and are based on the individual Market Lab project.
Grading thresholds
| Grade | Minimum Score | Additional Requirement |
|---|---|---|
| A+ A A- | 85 | Must be ranked in the top 30% |
| B+ B B- | 70 | — |
| C+ C C- | 60 | — |
| D | 50 | — |
| F | < 50 | — |
Subgrades such as A− or B+ are assigned based on the distribution of overall course grades.
Assignment
Apply ideas from 3073 to simulated trading in the Market Lab individual project. Your approach is open: traditional methods, LLMs, agents, and combinations are all welcome. Vibe coding is allowed.
- Project website and finance guide
- Project Requirements & Rubric
- Course starter code & data · Quick start
The starter includes cached StockBench data and a baseline that runs offline
without an API key or training. Submit an English LaTeX technical report
(main text at most 8 pages) and presentation slides on Canvas by
27 November 2026, 23:59 (UTC+8). Use NAME-STUDENTID for file names.
Each student gives a 20-minute presentation, followed by 5 minutes of discussion, in Lectures 12–13. The order will be published on Canvas. The project and presentation together contribute 30% of the course grade. Profit and loss amounts do not affect your grade. We assess your use of course ideas, technical depth, evidence, organization, visualization, and communication; see the linked rubric for details.
Last updated: October 22, 2026