Machine Learning Prelim

The purpose of the Machine Learning Preliminary Examination is to give students an opportunity to demonstrate their ability to analyze, evaluate, and present a pre-existing body of specific research in the area of machine learning. Examples of areas of research that fall within this field include, but are not limited to, the following topics: statistical learning algorithms, kernel methods, graphical models, Gaussian processes, dimensionality reduction and manifold learning, model selection, generalization, and Bayesian learning.  Included within the scope of the exam is application domains that rely heavily on machine learning methods, including but not limited to:  planning and control, time series prediction, bioinformatics, text/web analysis, robotics, game playing, computational modeling of human cognition and neural processes, speech and signal processing, machine vision, and image processing and coding. The specific topic area of the examination is expected to fall within within one of these areas.

The prelim consists of the following components:

Successful completion of this examination satisfies the Area Exam portion of the Computer Science department's Preliminary Examination requirement.

The review paper should summarize a minimum of 3 key papers in the chosen area, but more typically will address 5-8 papers.  Part of the challenge of the prelim is to pick a set of papers that are interrelated and form a coherent selection, and which can be sensibly compared and contrasted.  For example, the topic of unsupervised dimensionality reduction techniques might present, discuss, and compare recent techniques such as ISOMAP, LLE, and RBMs, and possibly to include for historical context classical techniques such as Kohonen maps.

The review paper should be double spaced with reasonable margins (e.g., 1" on every edge) and fonts (e.g., 10-12 point type).

Preparation

Typical preparation for the ML Prelim consists of CSCI 5622 (Machine Learning) and additional course, either CSCI 6622 (Advanced Machine Learning) or special topics courses that rely heavily on machine learniung approaches, such as CSCI 5832 (Natural Language Processing), CSCI 6302 (Speech Recognition and Synthesis), CSCI 7000 (Machine Vision),  CSCI 7782 (Cognitive Modeling), CSCI 7222 (Probabilistic Models). Other follow-on courses may be acceptable based on future course offerings and individual student concerns.

Note that while these courses are strongly recommended as preparation for the ML Prelim, students with transfer courses, or other kinds of preparation, may well be ready to take the prelim without taking these courses.

Logistics

Students will first identify a potential topic area, as well as a Computer Science faculty member who must agree to review and approve the topic.

The student is responsible for writing a brief (< 1 page) proposal that describes the topic area, specifies the approving faculty member, and proposes a minimal list of technical papers to be reviewed.  (We expect that the student will incorporate additional papers as the review paper is fleshed out.)  The approving faculty member will be available for consultation and consideration of papers, but it is the responsibility of the student to perform the background research necessary to delineate the topic and to identify the key papers in the area.

The brief proposal is submitted to the ML Prelim Chair for approval during the spring semester.  Once the Chair has approved both the topic area and the selected papers, the student has no more than 21 calendar days to prepare the review of the selected papers. The student may consult relevant faculty members with specific questions concerning the content of the individual papers, but cannot solicit or receive assistance of any kind on the overall analysis of the papers.

Copies of the completed review will then be delivered via both hard-copy and pdf to the department Graduate Advisor (Vicki Kunz) on or before the end of the 21 day period. Students may not submit preliminary drafts to any member of the committee for review.  The student is responsible for submitting a paper that has no grammatical or spelling errors.  Foreign students may ask a native English speaker to review the paper for grammatical corrections, but not for feedback on the content or presentation style.

The formal presentation will be held within four weeks of the submission of the paper. Given that three faculty schedules need to be coordinated, arrangements to schedule the date of the oral exam should begin as early in the semester as possible. Scheduling the presentation is the student's responsibility.

At the formal presentation students are expected to present the content of their review as they would at a technical conference. The committee's evaluation is based on the technical content, presentation style, and command of the area. Although fluency in English is not a requirement, students must be capable of clearly conveying the material orally. The presentation should be no more than 20 minutes in length. Students are strongly encouraged to make practice runs of their presentation to their peers, research associates, and faculty members who are not participating in the ML Prelim.

Miscellaneous

The subject area of the review paper may well correspond closely to a student's current area of research and planned thesis work. As such, it may overlap with a planned, or in progress, literature review section of a thesis proposal. This is explicitly permitted.

Prior papers written by the student—including conference papers, journal articles, masters theses, and class projects—cannot be submitted verbatim as a substitute for the ML Prelim paper. However, portions of such prior written work on which the student is the sole author may be re-used as the basis for part of the ML paper. Use of material where the student is one of several authors must be negotiated between the student and the ML faculty sponsor prior to the examination.

ML Faculty

At the current time, the core CS faculty members in the area of machine learning are: Gregory Grudic, Michael Mozer, Larry Hunter, and Robin Knight.  Professors Jane Mulligan, James Martin, Martha Palmer, Tim Brown, and Clayton Lewis have research interests that overlap with this area and in consultation with the ML Prelim Chair can approve ML prelim topics.

During the spring of 2008, Professor Gregory Grudic will serve as the chair of the ML prelim.