by Bill Freeman

ref: https://people.csail.mit.edu/billf/publications/How_To_Do_Research.pdf

---

Slow down to speed up

In research. No one tells you the right answer, we don’t know if there is a right answer

How to do it?

Verify your assumptions. Understand the thing, whatever it is–the program, the algorithm, or the proof.

Don't swing around doing random

you’ll be making much more progress than if you flail around, trying different things, but not understanding what’s going on.

Reading is part of the progress.

Progress can have many notions. One would be spending hours understanding what the problem actually is. (see love to hear about progress)

Don’t tell me “it doesn’t work”

Instead I like to hear:

“I’ve narrowed down the problem to step B. Until step A, you can see that it works, because you put in X and you get Y out, as we expect. You can see how it fails here at B. I’ve ruled out W and Z as the cause.”

Don't be like an average forum user

“I clicked run, it doesn’t work”. *Expand on it!

“This sounds like hard work.”

Everyone is smart. In graduate school, it’s the hard workers who pull ahead. This seems obvious, the harder question is “How do I get myself to work hard enough to do research well?”

“How do I get myself to work hard enough to do research well?”

work on that you can love. I think the Jaakko Lehtinen’s definition of doing PhD also simultaneously describes love (career-wise) for me too.

Success

If it solves something that people care, then I love it.

Steering

We need motors–that’s the working hard part. But we also need a rudder for steering–that’s stepping back periodically to make sure we’re working on the right thing.

Escape the rabbit hole

Balance in all things; go deep, but have a big picture.

Image courtesy of devpost.com

Time management books to be very helpful. They teach you how to spend your time solving the right problems.

• Research what is a good book that actually is not some bullshit that just says wake up 2 am, exercise for 15 hours, etc.

simplest toy model that captures the main idea

With a good one, you can build up intuition about what matters. Toy problems are advantages, because you can figure out what will work by thinking it through with your toy model.

Note

Also because the toy model is just the core idea without all the hassle, the readers (the main reason of effectiveness of a scientific publication) can grasp the idea more effectively. ⇒ you can even discuss the problem and core solution by elaborating the toy model in your manuscript.

Weak vs strong grad student

The strong student starts doing what the advisor has asked, sees that it doesn’t work, looks around within some epsilon ball of the original proposal to find what does work, and reports that solution.

I am a strong grad student

I entirely proposed the “super resolution” idea in the SCF21 paper. This is especially important that the first two times that I mentioned it (although I communicated it very badly), the idea got rejected. But, I spend many hours to properly formulate the idea, created an experimentation to evaluate the correctness of the assumption, and reported the output.

Sometimes it’s useful to think that everyone else is an idiot

This lets you do things that no one else is doing.

Warning

It’s best not to be too vocal about that

You have a brand: YOU

Build up a great reputation for yourself. E.g., does great work, have good ideas, great talks, write wonderful software.

I do great engineering

This is what I was complemented with many times

Cultivate your strengths and play to those strengths

E.g., being broad; creative; a great implementer; great at doing theory.

Success

I believe I am a great implementer.

Don’t report to me, “This instance doesn’t work”

Ask the following questions:

• Why doesn’t it work?
• Why should it work?
• Can you contort things to make an example that does work?
• Can you make it fail worse?

Warning

Expand, expand and expand on top of the problem. Don’t be like average pepega forum user; give us something to work with.

Danger

Don’t ask “attendance type question”: when you ask if “is there anyone who knows machine learning?” but you don’t ask the main question or what the problem is or how you prefer to be solved. The answer to your question is “yes I am”, but you lost your one chance.

E.g., in your research if you go to your supervisor and say “method X didn’t work”, your supervisor can say “ok, nice to know” and that’s it! Now go have fun with your research.

Love to hear about progress

Many PIs usually have a very general notion of progress.

• “I’ve shown why this doesn’t work”,
• “I’ve simplified the task to get it to start working.”
• “I spent the whole time reading because I know I have to understand this before I can make any progress.” (slow down to speed up)

Don’t hide from the PI

PIs like it when you track them down and insist that you need to talk. It is better to get unstuck now than see failure later on (failure == not progressing)

Most important quality for success in graduate school

Please see quality-for-success-in-graduate-school.

• I have Handbook_on_Time_Management_Skills.pdf but don’t know if it s good. (ref)

Love doing research

The research community is a community of people who are passionate about what they do, and we welcome you to it!