Skip to content

Start Here

Jump to bottom
Peter G. Chang edited this page Sep 1, 2024 · 8 revisions

Big Picture for the PhD

Goals for the PhD:

  1. Produce a thesis that you are really proud of and that prepares you for the next stage of your career (e.g., faculty job market).
  2. End up in a position where you are a leader in a sub-discipline.
  • Become a leader of a sub-field. Cultivate a much deeper understanding in your subject matter than most everybody else.
  • Develop the extraordinary background skills necessary to “feel” like an expert in your subject matter.
  • Develop enough depth of understanding to generate the new frontier of insights.
  1. Develop a sense of bigger picture for the team (in Sendhil’s lab) and an appreciation for what the bigger team is working on.
  • The research lab is currently focusing on really interesting things that are happening at the intersection of algorithms, models of people, and models of science.
  • Work in progress, but the current vision for the lab is: what do we think the future of computing looks like?
  • E.g., read Vannevar Bush’s 1954 "Some Things We Don't Know".
  • Develop a vision of the future that is more compatible and aligned with what we can realistically achieve.
  • Understand your relation to the broader mission of the group.
  1. Meta-cognitive goals.
  • Learn how to “think better”.
  • Cultivate a “taste” for research.

On Research "Taste"

  • Most people think that taste is something that is fixed and inherent.
  • But taste is not usually about articulating the first gut response, but it’s about cultivating taste.
  • We can cultivate taste by articulating it repeatedly with feedback.
  • At some point after a process of cultivation, you realize that your taste is something that is unique to yourself.
  • Need to be especially wary of the early instincts!

Creativity in Research

There are two kinds of know-how in research. Over time, the objective is that these two tracks of creativity in research will meet.

  1. The development and execution of an idea (e.g., producing a paper).
  • There is a lot of room for creativity here.
  • At some point, it will be a good practice to submit a paper in every conference cycle.
  1. The issue of taste and cultivating initial ideas.
  • Especially at the beginning of working on a research project, it is important to ask questions like why that project was picked or why other ideas may have been discarded.
  • Write down the ideas that you have in a notebook.

PhD Roadmap

Lab's Research Statement

We believe that the current factorization of disciplines is ill-suited for what we need for the next generation of cutting-edge research.

  • The path to the research frontier in the upcoming generation is not a simple linear one of mastering one specific sub-domain.

Core Skills

  1. Elements of computing
  • Optimization
  • Algorithms, data structures, and complexity
  • NLP
  • Computer vision
  • HCI
  1. Econometrics and statistics (theoretical)
  2. Empirical work (the "messiness" of empirical research)
  3. Psychology and behavioral sciences
  4. Engineering (the ability to build stuff)
  5. Know-how about interacting with the real world
  • Economics (principled deep understanding of the real world).
  • A practical set of know-hows around how to navigate the real world (e.g., RCTs).

Note that deep learning can be a subset of any of these core skills depending on the lens you look at it through.

Course Selection

When choosing which courses to take, differentiate between the three goals:

  1. Really understand (at an A+ level) and master two of the core skills in the list.
  • The means are more important than the ends.
  • The goal is to exercise the muscles required to really become an expert at an intuitive level.
  • E.g., if you take an economics course on some topic, the topic itself may be of secondary importance to the muscles built in the process of understanding the topic.
  • Try to take away timeless lessons from each of the course that you take.
  1. Exposure to new ideas.
  • Treat it like a buffet.
  • The level of care and intensity will be different to the first goal.
  • Swim upstream. Focus on learning the core way of thinking about a problem. Learn the fundamentals, aesthetics, and underlying structure of approaching problems in these domains.
  1. Achieve minimum competency for all the core skills.

The goal is to get to A+ mastery level in two core skills by the end of year two.

  • The motivation for this goal is that when faced with two powerful “incompatible” core skills at an intuitive level, there is an opportunity to create something new.
  • People in the CS department in general under-take classes.
  • The whole process is to figure out the “floor” of the minimum number of courses to take, and not the ceiling, so whatever else that you would like to add for fun, you can do so!
  • Most (young) CS people are really smart but lack this fundamental awareness of context outside of their narrow sub-field.

Clone this wiki locally