Syllabus

CS 598 FYY: Machine Learning for Systems is a graduate-level course at the University of Illinois Urbana-Champaign that explores how ML is changing the design and operation of computer systems and networks.

We will study and discuss a broad range of research papers, from seminal work that applies classical ML techniques to systems and networking problems to recent work that employs generative and agentic AI to design, configure, diagnose, and operate complex infrastructure. Through these papers, we will consider when learned approaches improve on traditional algorithms and heuristics, how they integrate with existing systems, and what is required for them to work reliably in practice.

Topics will span cloud and distributed systems, operating systems, database and storage systems, networked applications, mobile networks, and wide-area networks (including the Internet). Across these domains, we will examine applications such as scheduling, caching, configuration, congestion control, video bitrate adaptation, traffic engineering, anomaly detection, and failure diagnosis. In addition to performance, we will critically assess the robustness, generalizability, safety, and operational costs of deploying learned components in real systems and networks.

The instructor is committed to evolving the course based on student feedback and fostering a productive and inclusive learning environment. In Fall 2026, we are introducing mechanisms that promote effective teamwork while discouraging free riding. We will collect student feedback to evaluate how these mechanisms work in practice and to identify opportunities for improvement.

Grading

Class sessions

Class sessions focus on discussions of readings and student-led paper presentations, with additional background and commentary provided by the instructor. We will also invite external speakers to deliver guest lectures.

Active participation is essential to getting the most out of the class. You are strongly encouraged to read each assigned paper before class and come prepared to discuss it. As explained later, you may skip many paper reviews without affecting your grade; however, when you choose to skip a paper (and thereby save several hours that you would otherwise spend on it), you should make every effort to attend the corresponding presentation and interact with us.

As an incentive, 5% of the final grade (plus bonus points) is allocated to class participation, as described below.

Class participation

During each class, we will select several comments from the submitted paper reviews and invite their authors to elaborate on them. You may also volunteer to share your review if time permits. Whether you are invited or volunteer to do so, you will receive 1% toward your final grade.

In addition, if you participate in paper discussions during class (e.g., by asking or answering a technical question), you will also earn 1% toward your final grade. Additional opportunities to earn participation credit may be announced during the semester. In any case, log your participation as follows.

We typically don’t take attendance, so you are not required to attend every class — except for guest lectures and final presentations, as explained later. However, we believe that attending class is key to having a memorable and enjoyable semester! The ability to engage in technical discussions will serve you well in your future career.

If, at any point, class attendance becomes unacceptably low, we may have to introduce a new attendance policy that penalizes excessive absences. Any such policy will be announced on Campuswire.

Guest lectures

To respect the time of our guest speakers and ensure sufficient turnout, you are required to attend all guest lectures, unless you notify the course staff on Campuswire at least 24 hours in advance with a legitimate reason. If necessary, we may take attendance during guest lectures and deduct 1% from the final grade for each unexcused absence.

After attending a guest lecture, you will have the option to submit a brief reflection in the corresponding Campuswire assignment to receive 1% bonus credit toward the final grade. Your reflection only needs to be one or two paragraphs long. Rather than summarizing the lecture, it should focus on one or two insights, ideas, or results presented by the speaker that you found particularly interesting or surprising.

Readings and reviews

Each class has at most one required reading (as announced on the course schedule), allowing us to read, review, and discuss it in depth. Reading the additional readings is optional.

You are expected to study the required reading closely and seek out any background material you need to understand it. Depending on your familiarity with the topic, you can expect to spend anywhere from 3 to 8 hours studying each paper to fully comprehend the material.

Review requirements

For each required reading, you may submit a paper review containing (at least) three short comments, formatted as a numbered list with one comment per item. Your review should not summarize the paper or restate its abstract. Instead, each of your comments should (succinctly) provide information that is not found in the paper itself.

For example, a valid comment might be:

• a criticism or weakness of the paper;
• an advantage of the paper’s design that the authors did not discuss;
• a suggestion for extending or building on the paper in future work; or
• a connection to a previous or optional reading.

Review grading

Paper reviews account for 15% of the final grade, and you will receive 1.5% for each review that contains three valid comments, with each valid comment worth 0.5%. Thus, only 10 successful reviews are required to receive full credit, allowing you to skip many of the assigned papers if you prefer.

On the other hand, you are welcome to submit reviews for as many assigned papers (required readings only) as you wish. At the end of the semester, your 10 highest-scoring reviews will count toward your paper review grade.

This semester, we have intentionally lowered the number of reviews required to earn full credit. In return, we ask that you come up with each review and type it entirely on your own, without using AI to generate or polish any part of it. Do not worry about minor writing issues as long as your review is understandable.

We will grade your reviews leniently, but a review that sounds AI-generated or contains obvious hallucinations could result in an oral interview to assess your understanding of the relevant portion of the paper. If necessary, we may also conduct random oral interviews throughout the semester to screen for AI-generated reviews.

Paper presentation

You will be assigned to a small group of students to give one in-class presentation on a paper, much like presenting in an interactive reading group or seminar. Your primary goal is to help everyone gain a deeper understanding of the paper while fostering critical thinking and a lively discussion.

Presentation steps

1. Paper bidding: During the first week, you will receive a form to indicate your preferred topics from the course schedule. We will use your preferences to form presentation groups and assign papers. To facilitate communication, we will also create a private Campuswire chat room for each group to discuss logistical matters related to the presentation.
2. Presentation preparation: Work with your group to make slides and prepare the presentation. Your presentation should be self-contained and accessible to students who did not read the paper or found it difficult to understand. It should:
   • provide additional background on the concepts you found most challenging, going beyond the paper when necessary;
   • highlight results you found especially interesting or surprising, without attempting to explain every figure; and
   • engage the class by posing questions that prompt discussion of the paper’s design, results, or limitations.
3. Slide review: Submit a complete draft of your slide deck (or a link to it) to the course staff for review in your group’s chat room on Campuswire. The deadline is 11:59 pm three days before your presentation (i.e., Sun/Tue night for a Wed/Fri presentation, respectively). The slides do not need to be polished at this stage, but they must be in a “presentable” state. After receiving feedback from the course staff, you may continue revising the slides until your presentation.
4. Paper presentation: Prepare about 25 minutes of content if delivered without interruption, but you should engage the audience and take many questions throughout your presentation. You may use speaker notes or other supporting materials. The focus is on the quality of the technical content and the effectiveness of its delivery.
5. Posting slides: By the end of the day of your presentation, share the source file for your slides (e.g., a .pptx file or a link to Google Slides) in the Campuswire chat room. Also post a PDF version of your slides publicly on Campuswire so that other students can review them and ask follow-up questions.

Presentation grading

Your group may divide the presentation however you see fit, but every member must contribute to the preparation and present a meaningful portion of the paper in class. Any other special arrangements should be discussed with the instructor first.

The paper presentation constitutes 20% of the final grade:

• Individual contribution (5%). You will receive the full 5% as long as you make a meaningful contribution to the preparation. Specifically, immediately after your group submits the slides, briefly describe your contribution (in a sentence or two) in the Campuswire chat room, indicating what you were responsible for and whether you have completed it. This is a new component we are introducing this year to ensure that no one starts making slides the night before the presentation.
• Group presentation (15%). The remaining 15% is awarded based on the overall quality of the in-class presentation, including its structure and flow, technical accuracy and clarity, and your group’s handling of questions. To encourage teamwork, all members of the group will receive the same score for this component (except in exceptional circumstances). Feedback will be provided along with the score in the Campuswire chat room to help you improve your presentation skills.

Research project

The research project is the highlight of the course, accounting for 60% of the final grade. The goal is to conduct novel research that is broadly related to ML for systems and that, by the end of the semester, could lead to a publication-quality paper.

Your project should start with an important systems or networking problem and study whether “learning” (broadly speaking) can improve the state of the art. Refer to the project ideas page for more details. If you are uncertain whether your project fits the scope of the course, discuss it with the instructor first.

Although no paper submission or publication is required to complete the course, you are encouraged to keep a target venue in mind throughout the project. A practical target is one of the workshops listed under “ML for Systems” on the publication venues page, but any high-quality conference or workshop in systems and networking or AI/ML is also appropriate. Be aware, however, that publication norms can differ substantially across these communities.

The course materials also include pages on computing resources and research toolbox that are likely useful for your project.

Group logistics

Starting from the very first class, you should think about projects you might like to pursue. As you consider potential projects, find collaborators who are interested in the same topics. We recommend you work in groups of 3 to 5. If you wish to form a group of 2 or 6, you must obtain the instructor’s approval first; other group sizes are not allowed.

In any case, the deliverables should be roughly proportional to the group size; in particular, the final paper should be at least max(6, 2 × group size) pages long. In other words, the paper must be at least 6 pages, but for groups of 4, 5, and 6 students, the minimum length is 8, 10, and 12 pages, respectively.

For logistical matters related to the research project, we will create a (separate) private chat room on Campuswire for each project team. To seek technical feedback for your project, you are welcome to attend the course staff’s office hours.

Project milestones

0. Team formation: By the end of Week 2, submit your team roster using the form we will provide (each team only needs to submit once). If you cannot form a team by then, we will assign you to one.

1. Project proposal (5%): By the end of Week 3, submit a one-page research proposal that addresses each of the following points (not necessarily in this order):

   • Description of the problem you plan to address.
   • Importance of the problem.
   • Challenges of the problem.
   • Brief description of existing works (at least three) and why they fall short. Focus on high-quality papers published at top-tier venues and indicate which one represents the state of the art.
   • The goal of your project, especially the anticipated novelty and benefits of the potential solutions you are considering.
   • The initial steps you plan to take, along with a preliminary plan for dividing the work among team members.
   • References (excluded from the page limit).

   Use the provided LaTeX template for the proposal. Avoid exceeding one page (excluding references), but there is no hard page limit. The goal is to simply demonstrate that you have a plausible project and a general idea of how to approach it. You may continue to refine your idea throughout the project.

   Submit the proposal (PDF) in your team’s Campuswire chat room, along with the following statements (in plain text):

   • “Our team will adhere to the course AI policy and honor code. We will not engage in any form of academic misconduct.”
   • “Our team will meet at least once a week to discuss the project. Anyone who misses group meetings for two consecutive weeks must self-report to the course staff.”

2. Individual progress updates (15%): By the end of Weeks 6, 9, and 12, each team member must provide a short progress update (just a few paragraphs in plain text) in the Campuswire chat room.

   You will receive the full credit (5% per update) as long as you demonstrate meaningful progress, such as writing code, processing data, conducting experiments, implementing baselines, or developing theory. However, reading prior work (without analyzing or reproducing their results) or attending group meetings alone does not count as meaningful progress; you are expected to make other concrete progress beyond these activities.

3. Final presentation (15%): During Week 15, we will hold in-class project presentations. Attendance is required unless you notify the course staff on Campuswire at least 24 hours in advance with a legitimate reason. If necessary, we may take attendance during final presentations and deduct 1% from the final grade for each unexcused absence.

   The presentation should follow a conference format, i.e., an uninterrupted talk lasting ~10 minutes, followed by a 2-minute Q&A. The exact time limit and format will be announced on Campuswire. This is an opportunity for the course staff and other students to ask questions about your project and see the work you have accomplished!

   By the end of the day of your presentation, share the source file for your slides (e.g., a .pptx file or a link to Google Slides) in the Campuswire chat room. Also post a PDF version of your slides publicly on Campuswire so that other students can review them and ask follow-up questions.

4. Final paper (25%): By the last day of the final exam period, submit a publication-quality research paper (PDF) in the Campuswire chat room. Use our paper template (adapted from the USENIX LaTeX template). The minimum paper length is max(6, 2 × group size) pages.

Project FAQ

What kinds of projects meet the course expectations? In a nutshell, your project must produce something new. Novelty may come from a new system or learning method, a new combination of known techniques, a careful evaluation in a new setting, or the creation of a useful benchmark or dataset. A literature survey or reproduction of existing work alone does not satisfy our expectations. Even if the proposed approach does not achieve the intended outcome in the end, well-supported negative results might still count as a contribution for this course project.

Can my project overlap with independent research or another course project? It is acceptable to work on a class project that complements another ongoing project on a related topic. However, you should clearly identify a distinct component for this course project and request the instructor’s approval first. In any case, you may not reuse another project in its entirety. Be sure to check with your teammates and any other advisors or instructors involved as well.

Can I submit the final paper to a venue? If you plan to submit a paper largely based on your class project at some point, you should have an open conversation with your project collaborators, including anyone who provided substantive advice or assistance, to determine appropriate authorship. You are also welcome to consult the instructor for guidance on improving the manuscript and identifying suitable venues for submission. If your paper is accepted at a venue, please report back to the instructor so we can add it to our list of publications!

Course policies