Welcome to the Main Repository of quals-related material, maintained by your representatives in the PhD program committee. This is part of the Computer Science Department's PhD Bible. Here, we add information which is not detailed enough in the official program requirements [1], or represents cumulative knowledge of those who have come before us.
This is the most important (and time consuming) of the requirements for pre-candidacy students. It's also the most useful one. It is commonly admitted that the process of studying for quals (much more than the qual itself) gives students a better knowledge of their field of specialization, in a way that no other method could.
NOTE: A student who fails the qualifying exams twice is automatically dismissed from the department. In exceptional circumstances, the PhD Program Committee may accept a petition by the advisor and student to give the student another chance to pass the quals.
It's imperative (though optional) that you participate in a quals study group. Talking about what you study greatly helps you to establish what you've read, and form a more clear view of what is important and what isn't. Comparing notes with others on what the serious points are is a great cure against sidetracking or dilly-dallying. What's more, many PhDs cite their quals study group sessions as some of their most enjoyable experiences while in the program.
Below you'll find information on the different offered Quals per area: AI, Physiqual, Theory, Systems, Databases (InfoQual), and Biocomputation.
Below is a message from Professor Genesereth, the 2007-2008 Chair of the AI Qualification Exam:
The purpose of the AI Qualification Exam is to ensure that the successful candidate has in-depth knowledge of one or more substantial subareas of AI (usually the area or areas in which the student intends to do his or her doctoral research). Subareas include things like vision, robotics, probabilistic reasoning, computational logic, machine learning, multi-agent systems, and natural language processing (though there is no requirement that the subarea be one of these possibilities).
To pass this exam, the candidate must first assemble a qualification exam committee of three AI experts, of which at least two are active members of the Stanford Academic Council and affiliated with SAIL. The committee should be chaired by the student's research advisor.
Together with this committee, the candidate should agree upon the area or areas of AI that will be the focus of the exam. The candidate should then prepare a reading list of material in the selected areas, and the candidate should write a one-page explanation of the rationale behind this list. This explanation, along with the reading list, should be circulated to the committee and to the chair of the AI Qualifying Exam Committee for approval. The Chair of the AI Qualification Exam and the student's individual committee will then either approve the list or suggest further improvements to the candidate.
The AI Qualification Exam itself is an oral exam administered in one sitting. Two phases are recommended. In the first phase, the candidates should present the content of the papers in the reading list to the committee. Following the presentation, the committee should quiz the student about the topics covered by the reading list. A successful candidate must exhibit in-depth knowledge in the scientific areas covered by the reading list and must respond insightfully to the questions asked by the committee.
In evaluating the student's performance, the committee will consider three potential outcomes of the exam: Pass, Conditional Pass, and Fail. In the case of a conditional pass, the committee might place certain requirements on the student, such as taking or TAing classes. A failing candidate can retake this exam but has to begin the process from the very beginning.
Elaboration:
A couple of points about the qual - the first a clarification, the second a revision.
- SCOPE. Several of you have asked about the scope of the qual. Let me quote from the announcement:
"The purpose of the AI Qualification Exam is ensure that the successful candidate has in-depth knowledge of one or more substantial subareas of AI."
The phrase "substantial subarea" here means a "toplevel" subdivision of the field. Again from the announcement:
"Subareas include things like vision, robotics, probabilistic reasoning, computational logic, machine learning, multi-agent systems, and natural language processing (though there is no requirement that the subarea be one of these possibilities)."
Note that your committee may also require that you demonstrate mastery of several related substantial subareas, e.g. probilistic reasoning *and* machine learning. That is up to you and your subcommittee.
Your 1-page justification must identify the area or areas, and your reading list must realistically cover that area or those areas. Remember that your committee must approve your reading list, and so must I. Make sure it is of adequate scope or someone will complain.
A possible point of confusion: Your qual committee *may* ask you to read and critically analyze 2-3 papers as a way of assessing your knowledge of the associated AI subarea. However, the area is almost certain to be much broader; it is unlikely to be restricted to just 2-3 papers. Don't confuse the reading list defining your area with the set of papers that your committee asks you to focus on for purposes of discussion.
- SCHEDULE. Due to scheduling problems, some of you have asked whether it would be all right to take the qual after the May 20 deadline. In an act of rare and inexplicable flexibility, the faculty has decided to relent.
New rule: You can take the exam whenever you like so long as you meet the following constraints. (a) Departmental constraint: You may take the qual at most twice during your program. If you fail twice, you are out. (b) Departmental constraint: Normal progress guidelines say you should pass in your *second* year. However, if a student takes the qual in the second year and fails, we have generally allowed the student to try again in year 3. But that is it. If you wait till year three to take it for the first time and you fail, you are out. (c) Genesereth constraint: If you are planning to take the qual prior to the autumn quarter this year, you must tell me the expected date of your exam prior to May 20. Obviously, your committee must have agreed to this date or it doesn't count.
An older source on this qualifying exam is [2], Patrick Doyle's Quail Page. Although it hasn't been updated in a while, it contains lots of useful information, including paper summaries.
This qual covers vision, graphics, robotics, geometry and some applied math. For 2007-2008, Professor Fedkiw is in charge of this qual.
Information on the 2006-2007 physiqual is available at [10]; the information for 2007-2008 is likely to be very similar.
The Physiqual for the 2004-2005 academic year was held on April 4
from 9am to 4pm.
For 2001-2002, the physiqual page is at [7].
For 2007-2008, the faculty member in charge of the Theory qual is Professor Guibas. According to Professor Mitchell, the Algorithms qual format has been used for all theory topics for the past several years. Thus, the guidelines for all theory quals are below.
The examination procedure will be as follows. Each year an AA faculty member will be designated a quals coordinator. For each student taking the qual, the coordinator and student will work together to choose a 3-member committee. The committee, in consultation with the student, will select a topic. (Examples of topics are: approximation algorithms, data structures, primality testing, distributed graph algorithms.) Then the committee works with the student to select a set of (typically 3) papers in the chosen area.
The student is expected to read the papers and thoroughly understand the results and the proof techniques. To demonstrate this understanding to the committee, the student can come up with a unified framework that can be used to explain/prove the results presented in the papers and to show the relationship between them. Also, the student is expected to explain how these results mesh with the previous work in that area. Simplifying the arguments in the papers is desirable.
Sometimes a better framework leads to a better understanding which, in turn, leads to new publishable results. Although the student is not required to produce a publishable paper, producing such a paper essentially guarantees a passing grade.
The student has four weeks to prepare a report which is about 10 pages long. The oral part of the exam occurs within a week from the date the report is submitted. The oral part consists of a one hour presentation, open to the public, and a closed question session. During the closed session the student may be questioned on the material in the papers, as well as on background material related to the papers. The student is expected to be able to reconstruct proofs that were omitted from the papers.
During the oral part of the exam, the student may also be asked questions to test his/her general background. The student should consult with his/her committee to determine a readling list from which these questions may be drawn.
After the question session, the committee takes a vote to decide if the student passes or fails, and the student is immediately informed of the outcome. Afterwards, the student receives a letter with the feedback from the committee (e.g., keep up the good work, attend a specific course, improve writing skills). The committee also has an option to pass the student conditionally (e.g., subject to getting at least a ``B'' on a technical writing class).
The student is expected to work independently on all stages of the process. The student is allowed to discuss the material in the assigned papers only with the three committee members. Discussing the material with anyone else is forbidden; this includes discussion of the material in the papers, discussion of previous work, or discussion of different ways to present the material.
This exam can be taken at most two times.
The official page of the Analysis of Algorithms qual (from which the above material was drawn) is at [6].
For 2008-2009, the faculty member in charge is Professor Cheriton.
Dates for CS System Quals May 2009
The faculty in charge of the exams are as follows:
| Programming Languages | John Mitchell |
| Compilers | Monica Lam |
| Architecture | Kunle Olukotun |
| Networks and Distributed Systems | David Cheriton |
| Human Computer Interaction | Terry Winograd |
| Databases | Hector Garcia-Molina |
| Operating Systems | David Mazieres |
| Security | Dan Boneh |
| Graphics | Pat Hanrahan |
The following details applied to the 2009 portions of the Systems Qual:
The following details applied to the 2008 portions of the Systems Qual:
The official page for the Systems Qual used to be [3]
Some summaries for papers in the reading lists of the Operating Systems and Networking/Distributed Systems areas can be found in [4, 9].
Information on the 2008-2009 infoqual is available at http://infolab.stanford.edu/db_pages/infoqual.html.
Information on the 2007-2008 databases systems qual is below. Note that for 2008-2009, for the first component of the InfoQual (the Database portion of the Systems Qual) not all chapters are required if you use the second edition of the textbook. See the section on the Systems Qual above for more info.
For 2007-2008, the faculty member in charge of the databases qual was Professor Garcia-Molina.
The Database Group has a very good set of pages on the Database Qual [5]. The information below is taken from that site:
The InfoQual has three components:
- The Database portion of the Systems Qual.
- The InfoQual Oral: A 30-45 minute oral exam administered by Hector Garcia-Molina, Jennifer Widom, and one of Rajeev Motwani or Jeff Ullman. Material is drawn from the InfoQual Reading List.
- The InfoQual Written: The student and advisor together select 3-4 papers in a specific area of interest (e.g., the student's likely thesis area). The student writes a report of approximately 5 pages that summarizes, synthesizes, and criticizes the selected papers. The report is read by the advisor plus one more of Hector Garcia-Molina, Jennifer Widom, Rajeev Motwani, or Jeff Ullman.
Timing: Component 1 must be taken in the spring, at the time the Systems Qual is administered. Components 2 and 3 may be taken at any time, and there is no ordering requirement among the three components.
Passing: Components are passed/failed independently. Components 1 and 2 may be taken up to two times, and must be passed by the end of the third year. For component 3, the written report is expected to be satisfactory after one round of criticisms and suggested improvements from the two readers.
Report Quality: The quality of the written report will be evaluated on both content and presentation. Students are strongly encouraged to read through the very useful Elements of Style before beginning.
The following are guidelines for 2007-2008 provided by Professor Latombe, the faculty advisor for the biocomputation qual:
The Qualifying Examination tests a student's depth of knowledge and familiarity with the literature in his/her area of specialization.
The Biocomputation qualifying exam is a 1.5 to 2 hour oral exam in front of a committee of three faculty members, including the student's advisor. At least two members must have regular or courtesy appointments in the Computer Science Department. At least one must be from a "biology" department (e.g., Structural Biology, Chemistry, Radiology, etc...). In some cases, if appropriate, a committee member may be from outside the university. A committee member other than the advisor will chair the committee and will make sure that the exam is fair.
Three to five papers will have been selected by the committee one month before the exam. These papers will be related to the student's research area, but will not necessarily be specific to this area. For instance, if the student is interested in applying probabilistic learning techniques to predict gene functions, a paper may very well be about using similar techniques in understanding text in natural language. If the student is interested in molecular simulation, a paper on proximity maintenance among many moving objects may be assigned to the student. In general, the papers will be chosen so that the exam tests knowledge in computer science areas related to the student's research.
The student will be expected to have read these papers, as well as other related papers chosen by him/her. The choice of additional papers should be such that the student gains a good understanding of the topics covered by papers chosen by the committee. The student will indicate which papers he/she has read before the exam. The purpose of these papers is to both broaden the scope of the exam beyond the specific research interests of the student and limit the range of questions from the committee.
The oral exam will start with a student presentation of his/her research and a summary of the papers. This total presentation will last about 45 minutes. The rest of the exam will consist of questions asked by the committee and answers by the student. The questions will cover the student's research area, his/her readings, as well as broader computer science topics pertinent to the research area.
The exam will be set up specifically for each student. The student and his/her advisor will propose a committee for approval to the person designated by the CS chair to administer the Biocomputation quals (JCL in '07-08). Once approved, an exam date will be set up and the committee will select 3-5 papers for the student one month before this date.
At the end of the exam, the committee will deliberate and decide on the outcome of the exam. Three outcomes are possible:
In the latter case, the committee must specify the precise and verifiable condition that the student needs to satisfy, e.g., take a certain course and get at least an A , or TA a certain class, or write a 10-page report. A deadline for meeting the condition should also be specified. In all cases, a member of the committee other than the advisor will be mandated to verify that the condition is satisfied.
- Pass: the student has sufficient depth of knowledge and familiarity with the literature in his/her area of specialization to continue toward the PhD degree.
- Fail: the student has insufficient knowledge in his/her area of specialization.
- Conditional Pass: the student has few well-identified shortcomings in his/her knowledge of his/her area of specialization.
After the exam, the chair of the committee will send an email to the CS PhD program administrator (with copy to the person in charge of administering the Biocomp qual) indicating the outcome. If the outcome is a Conditional Pass, the condition should be clearly described. Once the condition is fulfilled by the student, the person who verified it should send an update email as well.
[2009/04/01] Updated for 2008-2009.
[2008/01/14] Updated for 2007-2008.
[2006/02/27] Updated for 2006-2007.
[2006/04/11] Updated for 2005-2006.
[2005/02/08] Updated the Physiqual section.
[2004/01/30] Updated for 2003-2004.
[2002/4/17] Updated link to Armando Fox's paper summaries.
[2001/10/11] Updated freshness and links for official quals pages.
[2001/02/14] Partially updated link to systems quals syllabus.
[2000/12/12] Added link to David Yu Chen's 2000-2001 study group.
[2000/10/31] Added link to updated official requirements page. Still no consistency checking between it and the bible.
[2000/6/19] Added a study group for Systems (George Candea's) for 2000-2001.
[2000/4/26] Updated with Systems Qual syllabus for May 2000.
[2000/1/7] Added Eran Segal's invite for an AI quals study group.