Improve on prior work: Problem pre-specified but solutions so far not
fully satisfying. E.g., algorithms & hardness results for classic
problems.
Develop new models:
Given phenomena or application that's not well
understood, develop new model/framework/criteria for understanding
key issues, explaining observations.
Working together on the board. Lots of
collaborative work -- not just with your advisor.
Bringing A and B together.
Lots of interesting relationships: ML - crypto - complexity - approximations
Ups and downs. Watch out for "obvious
but false"
Success, new insight -- feels great!
1. The people
Aesthetics
Problem Solving
Mathematical aptitude
A willingness to accept an imperfect model