PhD: My mistakes, my lucky breaks, and what I tell my students now

Thiago Serra
Assistant Professor, University of Iowa

Almost 8 years after defending my PhD, I am still figuring out what is the right way to pursue one. In retrospect, I was lucky to stay in academia given how I completed mine. I have also seen other students doing the right things, certainly more often than I did only to run out of luck at critical points. But now I must tell my own students something and, more importantly, direct them in a way that would help them succeed and hedge against that luck component. So let’s start with what I think I should have done, but which I didn’t.

The Thesis

When I start working with a PhD student, I tell them to think about where they want to be in 5 years, and then we work backwards from there. If they want to stay in academia, then the best outcome would be landing a tenure-track position that starts right after their PhD (or with a gap year, or two, to go for a postdoc in-between; but let us not chase unicorns here). In operations research, landing that position means being ready for the academic job market by early fall of their 5th year. But what does “being ready” even mean?

With a pool of hundreds of candidates for a tenure-track job, the first screening is often just a quick glance at their CVs. Are the papers there? Where were they submitted? Did they receive a first response at least? Thinking in terms of journal papers, we can work with the assumption of three good pieces of work, which look great on CVs. For instance, consider a scenario such as the following:

1. One paper published in a reputable journal, though not necessarily among the most prestigious venues, perhaps with substantial contribution from collaborators.
2. A second paper in advanced stage of review at a journal that is the best technical fit for the work, with the student having contributed significantly to all aspects.
3. A third paper under review and circulating as a working paper, which is largely the student’s own idea and demonstrates growing independence.

Every paper has a purpose. With the first paper, the student has experienced the full cycle of a research project, including resubmissions to the same or different journals, as well as dealing with reviews that were constructive and others that were unfair to their work. Learning that rejection is part of the parcel. With the second paper, the student signals the ability to produce strong work that is mostly their own. This paper is the most important because there is still enough time to reasonably expect to have a journal response prior to entering the job market. With the third paper, the student shows that they have established a research pipeline and that more work will follow in due time. By that stage, the student is ready to take off. So let’s talk about that backwards plan again, now in more detail:

1. To land a tenure-track job in five years, the student should ideally be ready for the market in four years.
2. But to have that second paper at least in major review, assuming an expected review time of six months and the occasional setback of rejection, whether for justified constructive feedback or unfair reviews, the paper likely needs to be ready within three years.

With that in mind, I came up with the following timeline:

1. Semesters 1 to 3: the student is primarily taking courses, while gradually ramping up research activities by contributing to existing projects and carrying out specific tasks. This is when the first paper is produced.
2. Semesters 4 to 6: having completed the coursework, the student becomes the main contributor of a project while relying on guidance from the advisor to identify a direction that is both productive and publishable. This leads to the second paper.
3. Semesters 7 to 9: if everything goes well, the student begins proposing the direction for the next step. This becomes the third paper, which will either appear on the CV as another submission (if completed by the end of semester 8) or as a working paper.
4. Semester 10: thesis defense, campus interviews, all the nail-biting that happens between interviews and hearing back, and pursuing a back-up plan.

In practice, however, projects defy our initial expectations of how long they would last. Sometimes we also get stuck, bored, or both. In addition, it is difficult to confine ourselves to a single line of work, especially when we are not entirely sure we want to commit to it in the long run. As a result, most students eventually want to have at least a second project going on. When my students reach this point, I have a second conversation with them. This is when I tell what I actually did, which was admittedly much messier and riskier.

The Antithesis

In the second conversation with my students, I show them what happened in my case. First, I go to my Google Scholar profile, sort the papers by year, and scroll down to the period in which I was a PhD student. There is some relevant work in 2012, the year before I started: my master’s thesis and a paper at the CP conference, which grew out of that thesis work. Then nothing else relevant until 2018, the year I defended: my PhD thesis and conference papers at ICAPS and ICML — and the ICML paper is not even part of my thesis work.

So what happened? I chose to work with two advisors on different topics, working on a main project with each of them as well as a spin-off project driven by my own questions about those projects. Then I pursued summer internships after my third and fourth years. Heading to my final year, I also started something entirely new on the side. That probably accounts for 80% of my PhD effort. I will skip the other 20%, which was even less focused.

For my PhD defense, all that I could show as evidence of external validation was the ICAPS paper from my fourth-year internship and a major revision at Mathematical Programming for one of my PhD projects, which had only been submitted shortly before I started my fifth year. I went to the job market a year later, already out of the PhD, and the only additional evidence was the ICML paper from that fifth-year exploration. Nevertheless, that was enough to get me a few interviews and eventually land a tenure-track position in 2019.

Continuing that conversation with my students, I go back to my Google Scholar profile and roll up to 2020, which I like to call the “miraculous year”: both projects with my advisors became journal papers, one spin-off also became a journal paper and the other became a conference paper. That fifth-year exploration yielded another two conference papers. Although none of that arrived in time to help me on the academic job market, the publication pattern becomes somewhat stable afterward — maybe I have learned my lesson!

In other words, I was very lucky to land a job before most of that materialized in 2020: every department that considered hiring me took a considerable leap of faith. They had probably seen similar cases before, and I appreciate their willingness to take the risk with me and others who similarly spread themselves too thin during their PhD years. However, the journey could have been much safer if I had completed some of that work earlier. But this entire discussion misses the main point: why is it that we do what we do?

The Synthesis

Every time I send a student away with objective steps for moving a project forward and the student comes back with questions and side explorations (and, of course, none of those objective steps implemented), I am reminded of why we do what we do: we are in this for the learning and the discovery. Indeed, if we do not pay attention, we could miss that a secondary issue to what we are exploring is actually a far more important problem to solve. In fact, it could be the best thing that we are ever do — but only if we notice it!

From that perspective, my choices were not so erratic as they may seem:

1. I chose to work with two advisors, John Hooker and Egon Balas, because I admired their work and wanted to learn from both. And I did learn a lot!
2. By pursuing those spin-offs, I learned to explore new ideas, had opportunities to fail, and developed a better sense of the kinds of problems I can — and should — tackle.
3. I chose to do internships because I wanted to continue thinking about how to connect my work with practice. That definitely helped ground my focus.
4. I returned to the same place for a second internship because I had a great manager, Arvind Raghunathan. He also left the door open for me to return as a one-year visitor after my PhD, which became a lifeline for my academic aspirations.
5. Through Arvind, I also met his friend Srikumar Ramalingam, who turned out to have a taste for problems that was very similar to mine. He eventually sent me a paper about deep learning, a topic which I kept ignoring because it seemed disconnected from my background. That paper sparked the fifth-year exploration with my PhD colleague and friend Christian Tjandraatmadja, which ultimately became the major theme of my research going forward.
6. Through Srikumar, I met his then student Xin Yu, who would lead the computational side of many of my projects involving deep learning in the following years. Together, we were able to do a lot more than we would have done on our own.

We do what we do because we love it. And the best way to do it well is by finding like-minded peers who become part of our success network. But without something tangible to show regularly as our contribution, we may put our long-term academic prospects in jeopardy. Hence, we need to have a plan — even if only as a benchmark against what we end up doing along the way.

Research is like cooking with too many pans on a two-burner stove top: leave a pan unattended for too long and its contents are lost; switch pans too frequently and nothing gets done; but stick with the same pans forever and you risk cooking something that will never become edible. So go ahead and flirt with another project if you must, but remember that there is only so much that you can cook before the guests arrive. And if the food is not ready, ill-mannered guests may take a bite out of you!

Note: I would like to thank Sabrina Harris, my partner in life, for many comments on how to improve this essay — and for bearing with me through the ups and downs of this journey!

Acknowledgements: We would like to thank Pavithra Sripathanallur Murali for taking the time to review this article.