Systems for Machine Learning

COMPSCI 692S (seminar)

Advances in large language models are reshaping how software systems are built, queried, optimized, and operated. Recent work on agentic AI, semantic query execution, and AI-assisted systems design suggests a future in which systems can interpret natural-language intent, call tools, orchestrate complex workflows, analyze data, and even help build or tune themselves. At the same time, these new capabilities raise important systems questions around reliability, performance, cost, correctness, evaluation, and integration with existing data and computing infrastructure.

This seminar will review recent research at the intersection of systems and agentic AI. Topics will include systems for agentic AI, semantic query execution over structured and unstructured data, LLM-based data processing, tool-using agents, and the use of agentic AI to build, optimize, or tune systems. The course will focus on reading, presenting, and critically discussing recent papers.

Prerequisites: Students are expected to have a strong background in graduate-level machine learning and in at least one systems area, such as operating systems, data systems, or systems for deep learning. Students in the 3-credit section are also expected to have strong programming skills and hands-on experience in one system area.

Class meetings: Wednesday at 1 PM, LGRC A104A. Attendance is mandatory.

Credits: Students may enroll for either 1 or 3 credits.

Syllabus: TBD.

Seminar structure

The course will consist of weekly meetings centered around student presentations and discussion of recent papers.

Students enrolled in the 1-credit section will complete the following:

  1. Paper presentation. Present one recent research paper from the reading list.
  2. Critical analysis. Identify a potential limitation, missing evaluation, or open question raised by the paper.

Students enrolled in the 3-credit section will complete the above and additionally work in a group to develop a small project around the paper. Each group will:

  1. Present a recent paper.
  2. Identify a potential limitation or open question.
  3. Propose an experimental evaluation plan to validate the limitation and explore possible solutions.
  4. Carry out the evaluation and present the results.