Meta-optimization: three lessons from the Ph.D.

Editor’s note: Rob Smith finished his OR Ph.D. in the spring, co-advised by me, Dr. Brandon McConnell, and Dr. Russell King. His work focuses on developing decision support tools and optimization models to support communications restoration after disruptions.

By Rob Smith

I recently provided my advisor with a write-up on what I learned during the Ph.D. process and what I would tell myself at the start. This write-up is a condensed version of that. Some of the recommendations I gave were obvious actions like “Pass Quals” or “Using classes to support research,” so I have omitted those in this piece. I believe the more interesting pieces of advice are, ironically, the “meta-optimization” portions—i.e., asking how we can get better at doing research in the optimization field.

First, I would highly recommend formulating as many hypotheses as possible. OR is not a lab science field. But the ability to consider what the model might generate helps you refine the solution or provides a neat storytelling point if it is not what you expected. This does not replace the process of formulating research questions, but it breaks down the process of answering them. The accelerative impact of doing this extends to many other areas of research, such as challenging assumptions in papers and identifying where topics apply across disciplines.

A good repo for all your research will benefit you, no matter where you want to go in the future. This should house all things: ideas, presentations, papers, and out-of-the-box things like failures. This is also where you can link the technical class notes to the work you are doing in the research. This setup can also be a place to store good papers, plots, and presentations that others have done for inspiration. The more you add, the more you will benefit in the long run.

The most nebulous recommendation concerns how to conduct research. I never really thought about this question while I was in graduate school. I do not think “doing research” looks like one particular thing. I think it is a summation of many skills you learn to do well and with confidence, even when the path forward is ambiguous. These actions are: conveying ideas (that you may only understand), presenting ideas and plots, identifying roadblocks in your research, making assumptions, reading the literature, connecting ideas, and building models. Research is highly non-linear and uncertain. Being able to complete the immediate task in front of you and connect it to its destination is the finite unit of this training and experience.

The research process and, thereby, graduate school are different for everyone, but hopefully these three ideas are helpful. Things have changed immensely since I started, given the rise of generative AI. But these three skills can all be done with pen and paper, so hopefully that gives them longevity.