Systems for Machine Learning

COMPSCI 692S (seminar)

Advances in machine learning (ML) and deep learning are constantly transforming prototypes in research labs to valid solutions to real-world problems. Using ML entails developing end-to-end pipelines to collect data, preprocess it, and run learning and inference algorithms in a scalable manner. This results in computationally intensive workloads and complex software pipelines. Systems for ML help users organize their data and scale these computationally intensive problems to larger and larger datasets. This seminar will review cutting-edge research on these topics. It will focus on reading, presenting, and discussing recent papers in the domain of ML for systems (1 credit). The instructor will offer some 3-credit follow-up independent studies.

Prerequisites: None. Background on ML and deep neural networks (COMPSCI 589, 689, 682, or similar) is strongly recommended.

Class meetings: TBA.

Credits: 1 credit. The instructor will offer some 3-credit follow-up independent studies.

Schedule and syllabus: TBA.

Seminar structure

The course will consist of meetings with presentations. Students will be expected to participate in the following activities.

  • Presentations. Each student will have to present, alone or in a group, at one of the meetings. The presentation will cover one paper taken from a reading list published on Moodle. Each presentation will have 3 parts:
    1. Background and motivation. Present the general topic addressed in this paper, the prior related work, and the gaps that this paper addresses. Presenters are encouraged to read the most relevant work in the area to prepare this part of the presentation. (~15 minutes)
    2. Paper content. Presentation of the technical content of the paper. (~15 minutes)
    3. Potential extensions. Propose potential research directions extending the work. Optionally, this section can be presented as a specific project proposal with expected goals and intermediate milestones. (~10 minutes)
      • The instructor will offer some 3-credit independent studies based on well-defined project proposals.
    4. Discussion. Questions about the paper and discussion on the general area, the paper contributions, and the potential extensions. (open ended)

  • Paper reviews. Two days prior to each presentation, each student will have to read the presented paper and enter a review on a Google form. After the deadline, reviews will be published to the class to encourage discussions. Reviews will also be discussed in class.

  • Attendance. Attendance to the presentations is mandatory.