Fall 2026
Lecture: MW 10:30 - 12
185 EWRE
Personal data about individuals is being collected at an unprecedented rate today. On one hand, analyzing this data uncovers new insights and functionalities that improve downstream applications. On the other hand, the collected data is highly sensitive and raises serious privacy concerns. This creates a constant tension between data discovery and data privacy.
This course will serve as a primer on developing full-stack privacy solutions – we will explore why personal data collection and analysis is considered a privacy risk, what constitutes a good privacy definition, and how to achieve privacy-preserving data analysis in practice. We will cover beautiful mathematical concepts from machine learning, differential privacy, and cryptography, and explore how these techniques are used to solve real-world privacy problems. For instance, how can Google train a model for next-word prediction for the Gboard app without accessing data on our mobile phones? How does the US Census Bureau release statistics without revealing any information about individuals? How can a bank prove the fairness of the model it uses for loan approvals without disclosing the model weights to anyone?
A particular focus of the material will be the recent revolution in large language models (LLMs) and the unique privacy challenges of LLMs and agentic systems.
By the end of the course, my hope is that you will have a deeper understanding of the privacy challenges we are faced with today and be better equipped to solve them.
Prerequisites
EECS 498: (EECS 388 or EECS 445) or (senior standing with major in CSE or Data Science or Mathematics or Statistics) or permission of the instructor.
EECS 598: The course requires graduate standing. While there are no formal requirements, the course will teach techniques from cryptography, machine learning, algorithms, and probability.
If you do not have the formal prerequisites listed above but are motivated to take the class, please feel free to reach out to me.
Structure
This is a research seminar class. It will consist of instructor lectures and student-led research paper discussions. Evaluation will be based on
class participation (20%)
paper presentations (30%)
an open-ended semester-long team research project (50%)
The research project can be quite flexible: it could be entirely implementation-based, purely theoretical, or a blend of both.
Every student needs to do (at least) one presentation. Since discussing the research is a large part of how students learn about the material, you must attend class in person.
Your class participation will be graded both on your class attendance and how much you actually participate in class discussions. One good way to participate is to ask the presenter thoughtful questions about the research.
Your paper presentation will be graded on how well you explain the research and answer questions. The extent to which you actually understand the material you are tasked with presenting will be judged in large part by your ability to answer questions during your presentation, both from other students and from the instructor. If you cannot answer questions about your assigned material satisfactorily - for example, because an LLM wrote your presentation and you didn't really read the papers - your grade will be negatively impacted. See the Grading page for more detailed information about how presentations are graded.