This course explores the intersection of AI and systems with a focus on how AI techniques can help design, optimize, and manage computer systems. Modern systems are facing growing complexity that arises from scale, heterogeneous hardware, deep dependencies, and highly dynamic workloads. These complexities make it difficult to rely on traditional approaches to reason about, control, and predict the system behavior. At the same time, as the backbone of computing services, systems today face ever more stringent requirements about their reliability, availability, performance, security, etc. The advances of AI techniques provide opportunities to address these gaps.
The course covers a broad range of systems, including operating systems, distributed systems, databases, storage, and networking. It examines how AI techniques can be applied to key system components such as schedulers, file systems, caching and memory management, query optimizers, and distributed protocols. Rather than focusing solely on parameter tuning, the course will take a holistic view of how learning-based approaches can deeply integrate with complex systems to guide system design, implementation, policy decisions, and augment or rethink traditional system mechanisms.
The course traces the evolution of AI for systems research. It begins with early work that predates deep learning, such as statistical modeling and classical machine learning for system optimization. It then progresses to more recent advances, such as learned system components and leveraging large language models (LLMs) for system-related tasks.
Scope: This course focuses on AI for Systems: using AI techniques to design, optimize, and manage computer systems. It is NOT about Systems for AI, which studies systems and infrastructure for supporting AI workloads.
By the end of the course, students will be able to:
Prerequisite: EECS 482 (Introduction to Operating Systems) or equivalent systems background. Prior exposure to machine learning (e.g., EECS 445) is helpful but not required.
| Lecture | Tuesday/Thursday 3:00pm-4:30pm, 1200 EECS |
| First class | Tuesday, September 1, 2026 |
| Last class | Thursday, December 10, 2026 |
This course is research oriented. It is structured as a mixture of instructor-led lectures, student-led paper presentations, guest lectures and discussions. There are very few traditional lectures; most class meetings are discussions of research papers that everyone is expected to have read beforehand.
This course does not have a required textbook. The course readings consist primarily of research papers from top systems and interdisciplinary venues such as SOSP, OSDI, NSDI, SIGCOMM, SIGMOD/VLDB, ASPLOS, and MLSys. These papers include both foundational work and recent progress so students understand the evolution of ideas and recurring challenges.
Most classes will have two required papers to read. You MUST read the required papers before class and be prepared to discuss them. For some topics, we will list additional optional papers. You are encouraged, but not required, to read those. Students often find it useful to form a reading group to discuss papers together before the class period, and we encourage the practice.
If your machine learning background is rusty, the following references are useful companions while reading the papers. They are not required:
Note: this course assumes that you have taken an undergraduate operating systems course. It will not cover background knowledge such as how virtual memory or a scheduler works, and most papers we read make the same assumption. If your undergraduate systems knowledge is rusty, reviewing past lectures (example) will help you better comprehend the papers.
For the group-based projects, each student is expected to make substantial and clearly identifiable technical contributions and will be assessed based on their contributions. To support fair evaluation, students will be asked to submit individual contribution summaries at major milestones. Significant imbalance in contribution will result in different grades among members of the same team.
Participation: Participation grades are based on consistent engagement with course discussions and activities rather than speaking frequency alone. Students are evaluated using multiple indicators, including:
Each student will also serve as a discussion leader for assigned papers during the semester, which contributes to the participation component. If you remain silent in the class, you will gain little out of this course.
Paper Review: You are required to submit a brief review for the required reading in each class. Reviews must be turned in by 12 pm on the day of the class. Please refer to the detailed requirements on this page.
Presentation: Each student will be responsible for presenting and leading discussion for one or more research papers during the semester. The goal of the presentation is not only to summarize the paper, but also to critically analyze its motivation, design choices, methodology, and implications. Presenters are expected to discuss with the instructor in advance to ensure appropriate depth and scope. Detailed requirements are described on this page.
Attendance: Three absences of classes are allowed for good reasons (e.g., conference presentation, interview). However, reviews of the papers must still be turned in.
Discussion: You are strongly encouraged to discuss the papers with other students in the class – you may have insights that others do not, and vice versa. Often students form reading groups, which I heartily encourage. Note that group discussion, however, is not an effective substitute for actually reading the paper.
Project: A major component of this course is a semester-long, research-oriented project on using AI techniques to address a systems problem. You will work in a team of 2-3 students. Detailed requirements for the project are described on this page.
Missing and Late Submission: You may miss up to three paper reviews without penalty. For each additional missed review beyond three, a 30% deduction will be applied. Therefore, missing a total of seven paper reviews results in a zero for the review grade.
Missing the submission of any other assignment (e.g., proposal) will result in a zero for that assignment. Late submissions of any assignment (including paper reviews, proposals, etc.) will incur the following penalties:
Academic Integrity: All students are required to know and adhere to the College of Engineering Honor Code and university policies. They apply to all course requirements, including the paper reviews, proposals, course project code and documentation. Violations of the policies will result in serious consequences.
Use of GenAI: This is a course about AI techniques, and you will read papers and build projects that use LLMs. That said, you may not use GenAI tools to do the required intellectual work for you, such as generating your paper reviews, answering the assigned paper questions, or writing your project reports. You may use them to help you understand background and concepts in a paper, and as a coding aid in your project. In the latter case, you must disclose how they were used in your write-up.