University of Southern California

T-Troup

What is T-Troup?:

F-Troup is a misspelling of the title of the 1960s US television show "F-Troop", created by my Ph.D. advisor David J. Farber. The list refers to his group of advisees (mostly current, but some former as well). The "F" in F-Troup refers to Farber.

Some of Dave's students carry on this tradition in their own groups, using either their first or last name for the initial letter. This has also propagated to his students' students:

G-Troup (Guru Parulkar)

T-Troup (Joe Touch)

A Word About Lists:

There is a key point about lists worth noting before you see too many, e.g., below. Lists are very important to a Ph.D. They show your ability to organize, prioritize, condense, and express information. Some hints:

Parallel concepts: if you list 4 things, they should be equally important and similar in scope, i.e., parallel in size, structure, etc.

Necessary: if you list 4 things, are all really required, or can you do with just 3?

Sufficient: if you list 4 things, are they enough, or would someone else need to add a 5th, or a 6th?

Here are three things committees are looking for in a list:

A list shows you have mastered factual correctness and completeness, which ensures sure you didn't omit a whole topic area

A list shows you've added some knowledge to what's out there. This is what most students think earns them their PhD.

  • A. Mastery of knowledge
  • B. Accomplishment
  • C. Judgement

This shows you know what's important and what isn't. This is what most advisors think earns a student their PhD

Lists are about C. C determines when you are ready to graduate, more than A and B combined, IMO.

7 Questions:

Getting OUT of graduate school is the primary job of every graduate student. Every other force keeps them in. There are 7 questions that are the acid test for graduating.

  • Answer the first 4, and you have a topic to focus on.
  • Answer all 7, and you're ready to write your proposal.
  • Satisfy your answer to the 7th question, and you're ready to write your dissertation and graduate!

The following are my list of 7 questions:

  • T1: What is it?
  • T2: Who cares?
  • T3: Has it been done before?
  • T4: Why you?
  • T5: Is this worth a PhD, no more, no less?
  • T6: What new areas does this open?
  • T7: What is the objective metric where you, me, and your committee know you are done?

NOTE: all questions except T3 and T7 should be answered in one simple, direct sentence.

Some have noted the similarity between this list and the infamous DARPA Heilmeier criteria for new research programs, which are:

  • H1: What are you trying to do? (T1)
  • H2: How is it done today? (T3) What are the limitations of the current practice? (T2)
  • H3: What is new in your approach and why do you think it can succeed? (T4)
  • H4: Assuming you are successful, what difference does it make? (T2, T6)
  • H5: How long will it take? (T5) How much will it cost? (T5) What are the midterm and final exams? (T7)

Writing papers:

Writing is one of the more important tasks for researchers, and one for which typical college-prep and college education is poor preparation. Why?

Because high-school and college English classses tend to teach students to write like English writers, writing short stories and things that have the flavor of novels. Unfortunately, science is not amenable to suspense, erudite composition, or complex semantics.

What did I just say? Exactly! Science writing bears little similarity to most books studied in English classes, and much more to newspapers.Why? You're reporting on your research.

So what does that mean? Simple:

  • write in simple, direct English
  • lead with the main point
  • make all other ideas back that point up

Apply these rules at all levels of structure, from the chapter outline down to the paragraph.

See in particular: http://www.eecs.harvard.edu/~htk/thesis.htm

Here's the basic outline that many good research papers follow:

  • Intro
    • state the problem (this is WHY)
    • state your solution (this is WHAT, not HOW - HOW comes later)
    • state the benefit of your solution
    • explain the structure of the rest of the paper
  • prior work
    • group into categories; for each set, describe what they do and how you're different
    • summarize how you're different from each set and summarize your differences
  • your solution
    • explain in detail; this is HOW
  • your evidence
    • explain the experiments
    • describe the results - but not what they mean (yet)
  • discussion
    • interpret the experiments - what do they mean?
    • add some insight - read beyond the data a little
    • what is the impact of all of this work?
    • give some pros and cons to your solution, your evidence, etc.
    • discuss some possible future work and open questions
  • summary
    • highlight the overall paper briefly, i.e., one sentence per section or so
    • restate the overall solution, the reason you believe it, and its impact

Giving talks:

The purpose of a talk is to sell the audience on going to your website, reading your papers, etc. It is NOT to teach them the content of those papers!

Just as good writing is based on newspaper articles, good talks are based on newspaper cartoons. You get a few panes, and a few words per pane. Think Dilbert or Snoopy, not Prince Valiant or Dick Tracy.

Hints:

  • never more than 5 items, nor less than 2 in any list
  • never more than 5 words on a line (this isn't prose!)
  • do NOT read your slides; your audience can read already. talk _about_ your slides.
  • a picture is worth a lot more than you think, but SIMPLE pictures are the point. Again, _cartoon_ your concept
  • do NOT put code on a slide (at best, only pseudocode should fit anyway)

Writing programs:

  1. Functions should fit on one or two pages. If they do not, you have not properly decomposed your problem.
  2. Functions should have not more than 6-7 arguments. (see 1.).
  3. Numeric/alphabetic constants should not appear within your code. Use the DEFINE statement of the preprocessor to 'declare' constants.
  4. Indent your code by some uniform scheme; use automated formatting programs for this purpose, if you prefer. (cb file )
  5. Comments and written explaination of user procedures are considered part of your code, and integeral to its use. Document your assignments accordingly.
  6. Prefer readability to cleverness in your code. Remember, we use high-level languages for other people to follow. Your goal is not speed; if it were, you'd be better off using assembler languages.
  7. When in doubt, remember that a day of planning is worth a week of debugging.

Writing letters of recommendation:

Students become employers or teachers (hopefully). So here's a suggested outline of a recommendation letter, to help you provide both the information and the context many employers are expecting:

  1. introduce yourself - why we should believe your judgement, in a general sense
    • who you are
    • what you do
    • how many students/employees do you work with
    • how many years have you been doing this
  2. explain how you know the person you're writing about - why we should believe what you say about *this* person?
    • when/how did you meet
    • why did you hire/accept them
    • what did you ask them to do
    • what did they do for you (just the facts here)
  3. evaluate the person's work with you - (evaluate past performance)
    • attitude
    • responsibility
    • competence
    • do they follow directions/instruction
    • are they a self-starter
    • are they a self-learner
    • how well they interacted with others
    • how well they interacted with you
    • (include both good and bad - some weak points are OK, and quite useful; they help make the recommendation believable)
  4. evaluate how the person is suited for what they're being recommended to - (evaluate future potential)
    • why is this person a good fit
    • why is this person worth a chance
    • what will they add
    • what should you look out for to help with
    • what could be nurtured/expanded

Conflict of Interest:

Conflict-of-interest (COI) is a widespread concept, but often poorly understood. It is, IMO, based on "the POTENTIAL APPEARANCE of INAPPROPRIATE BIAS". Each word in that phrase is important. We can evolve the concept by expanding it incrementally:

  • BIAS: everyone (I hope) knows what bias means; this means you have a preference for a particular viewpoint or opinion. Some biases are based on evidence, others on experience or intuition. Bias by itself is not bad.
  • INAPPROPRIATE BIAS: We all know that COI is related to an inappropriate bias, and some forms are obvious. E.g., if you are bribed to give a good review, that would be inappropriate bias. In that case, the inappropriatness is real, and those are the easy cases. But that's not all there is to COI.
  • An APPEARANCE of INAPPROPRIATE BIAS: It turns out not to matter what you do, so much as what others think you did. For example, you give your kid a job. That kid might be the most qualified person for the job, but you're their parent, and even a reasonable person might think you hired them solely because of that relation.
  • The POTENTIAL APPEARANCE of INAPPROPRIATE BIAS: It doesn't even matter whether you think someone else feels that you did something wrong. What matters is that *someone* might. A reasonable person - i.e., we're not assuming that anyone on  the entire planet might possibly think, but what we can generally expect might happen.

So, in summary, it's not about what you do. It's not about whether it was actually OK. It's not even about whether it looked OK. It's whether others MIGHT think it wasn't.

So it's fine to think that a technology is bad, but if you hate flourescent lights because you think they're evil, maybe you should excuse yourself from reviewing papers about flourescent lights - pro or con.

T-Speak:

I have a habit, formed in my pre-college years, of over-quoting movies in conversation. This works fine for students who share my US, pop-culture-laden upbringing, but can be a challenge for others.

The following movies are suggested viewing for T-Troup members for that purpose. Note that some movies are a series; in general, the second movie should not exist, and so should be ignored:

  • Comedy
    • Real Genius
    • LA Story (esp. if you visit LA)
    • Raising Arizona
    • My Blue Heaven
    • Monty Python and the Holy Grail
    • A Christmas Story
    • Joe vs. the Volcano
    • The History of the World, Part I
    • Caddyshack
    • Stripes
  • Action
    • Raiders of the Lost Ark (I and III only)
    • Hunt for Red October
    • Goodfellas
    • Streetfighter
    • The Great Escape
  • Sci-Fi
    • Back to the Future (I and III only)
    • 2001: A Space Odyssey
    • Ghostbusters (I only)
    • Buckaroo Banzai
  • Drama
    • The Princess Bride
    • The Hudsucker Proxy
    • Bridge on the River Kwai
    • Citizen Kane
Groups: