Data Privacy (Fall 2026)
Course overview
This course asks how we can build useful data and AI systems without exposing
the people behind them. For three real privacy failures and the defenses that
answer them, see Why take this course?.
Who should enroll?
This course is open to all undergraduates.
Required preparation:
- Completion of an introductory programming course plus at least one 2000-level
computing course, or equivalent programming experience.
- Basic probability and statistics, including distributions, averages, and
simple evaluation metrics.
- Be comfortable working with AI assistants to inspect code, debug, and
organize an investigation while checking their claims against code, data,
and tests.
Helpful UVA courses, but not prerequisites:
- CS 3710: Introduction to Cybersecurity, for threat models, attacks, and
defenses.
- CS 4774: Machine Learning, for model training and evaluation.
No prior coursework in privacy, cybersecurity, machine learning, cryptography,
or AI is required.
The optional readiness self-check gives a concrete picture of
the programming, probability, and ML ideas used at the start of the course. It
is ungraded and is not an enrollment requirement.
Course info
- Course: CS 4501-003, Special Topics in Computer Science: Data Privacy
- Instructor: Tianhao Wang
- Email: tianhao@virginia.edu
- Location: Olsson Hall 011
- Time: Tuesdays and Thursdays, 2:00-3:15 PM
- Dates: August 25-December 8, 2026
- Canvas: course site
- Format: in-person lectures, short practicals, four varied labs, quizzes, Canvas check-ins, and a scoped final project
- Syllabus: Fall 2026 syllabus
- Policies: course policies
Announcements and assignment updates appear in Canvas. Use Canvas Inbox or
email for questions that involve grades, accommodations, or other private
matters.
Grading
- Labs (40%): four individual, equally weighted challenges covering privacy attacks, a two-agent secret arena, hidden tests, extensions to large DP codebases, and privacy-enhancing technologies.
- Project (30%): an innovative privacy system, tool, attack, defense, monitor/auditor, or research contribution, with an in-class project pitch (10%), proposal (5%), poster/demo (10%), and final report (5%). See the milestone guide and project rubric.
- Quizzes (20%): two individual, in-class, closed-book paper quizzes worth 10% each.
- Exit surveys / check-ins (10%): short individual Canvas submissions completed across the semester for completion credit.
- Grading basis: we use the UVA grading basis; letter-grade thresholds will appear in the syllabus.
Schedule
Week
Tue
Thu
Milestones
01Aug 24-28
Privacy definitions and attack taxonomy
Readiness check
02Aug 31-Sep 4
Extraction and memorization
Membership inference and attack evaluation
03Sep 7-11
Security notions and adversarial ML overview
ML security attacks and LLM agent privacy surfaces
Lab 1 out
04Sep 14-18
Anonymization, linkage, and k-anonymity
Differential privacy: definitions and basic mechanisms
05Sep 21-25
Laplace and Gaussian mechanisms
Composition, selection, and sparse vector
Lab 1 due · Lab 2 out
06Sep 28-Oct 2
Private learning: DP-SGD and PATE
Advanced and local differential privacy
07Oct 5-9
No class: Fall Reading Days
Project pitch presentations
Project pitch
08Oct 12-16
Guest lecture: TEEs and confidential LLM serving
Quiz 1
09Oct 19-23
Introduction to privacy-enhancing technologies
Symmetric-key cryptography, hashes, and MACs
Lab 2 due · Lab 3 out
10Oct 26-30
Public-key cryptography and Diffie-Hellman
RSA, digital signatures, and timing attacks
Project proposal
11Nov 2-6
Secure multiparty computation
12Nov 9-13
MPC: secret sharing, malicious security, and oblivious transfer
Homomorphic encryption, zero knowledge, and verifiable computation
Lab 3 due · Lab 4 out
13Nov 16-20
Guest lecture: oblivious RAM (ORAM)
Quiz 2
14Nov 23-27
Network privacy and contextual integrity
No class: Thanksgiving recess
15Nov 30-Dec 4
Guest lecture: privacy in industry
Guest lecture: emerging privacy research
Lab 4 due
16Dec 7-11
Poster/demo session and course recap
Poster/demo · Final report
More resources
AI access
- Gemini and NotebookLM through UVA
are available to UVA faculty, staff, and students at no additional student
cost.
- UVA Copilot Chat is also
university-licensed and available at no additional student cost.
- Eligible U.S. college students can claim
one year of Google AI Pro at no cost and
four free months of ChatGPT Plus. The
offers must be claimed by December 31 and October 31, 2026, respectively;
both require student verification and a payment method and become paid
subscriptions unless canceled.
- UVA RC GenAI provides
Kimi K2.5 through a browser and API at no charge to eligible Research
Computing users. RC currently restricts this service to research use; it is
not available for ordinary class assignments.
Courses
Core differential privacy
Broader privacy, systems, and machine learning
Tutorials and practical guidance
Software
Lab and course libraries
- Opacus for private training in PyTorch.
- JAX Privacy for auditable DP training in JAX and Keras.
- Google DPSynth for differentially private tabular synthesis.
- MP-SPDZ for secure multi-party computation.
Additional libraries
- OpenDP for composable DP mechanisms and data analysis in Python, R, and Rust.
- Tumult Analytics for scalable DP queries over tabular data.
- Google Differential Privacy for production-oriented C++, Go, Java, and data-pipeline components.
- diffprivlib for scikit-learn-style private models, mechanisms, and accounting in Python.
Books
Cryptography & MPC
Differential privacy