I recently moved from the University of North Carolina at Chapel Hill to the University of Toronto, where I'm now an associate professor and an associate research director at the Vector Institute. The path to this move was long, winding, and occasionally unconventional, so I'm writing this blog post in hopes that it's interesting or useful for others with a similar path.

The not-postdoc

My goal in pursuing a PhD was to eventually pursue a professorship. I knew that I wanted to continue to do research and I also had the sense that I enjoyed mentorship and teaching. I did my PhD at the intersection of music and machine learning, but by the end of my PhD I was primarily interested in machine learning itself. I also had a sense that, while I had produced some successful work during my PhD, most universities are not desperate to hire people who work on music and I also hadn't made enough of an impact to have assured success on the faculty job market.

So, I figured I should try to do a postdoc to gain more experience doing “core” machine learning research. I finished my PhD in 2016, when companies were aggressively hiring researchers with deep learning experience, and consequently professors were desperate to hire postdocs - desperate enough that Yoshua Bengio was willing to make me, with my weird music background, a postdoc offer at MILA. In the meantime, I had heard about the Google Brain Residency program (which was in its first year and had been pitched to me as “sort of a postdoc that doesn't require a PhD”) and ended up applying to it. I was lucky enough to get into the Brain Residency too, and ultimately decided to do the residency for a year and then consider a postdoc after (and MILA was fortunately willing to consider this). Why do the residency? The biggest factor was probably the fact that my girlfriend at the time (now my wife) didn't speak French, and it seemed like it would have been challenging for her to get a job in her field in Montreal. In any event, after completing the year-long residency, I decided to stay as a full-time research scientist (which incidentally is the subject of the first blog post I ever wrote), but my intention was always to apply for faculty positions eventually.

The first round

It took me a few years to start to feel like I had built up a solid portfolio of machine learning research that could be the basis of a faculty job application. During that time, I started to think that I might have an ok chance on the market - in particular, the strong “buyer's market” for deep learning researchers had continued and many schools seemed to be highly motivated to hire in the field that I was increasingly a part of. In addition, I started to get some encouraging signs - people telling me they thought I'd do well on the market and/or encouraging me to apply to their school. So, in 2018, I applied to around 25 schools, which — for the sake of brevity — let's just say were the intersection of the top 50 schools in CS (according to whatever faulty-but-not-totally-wrong ranking you want to use) intersected with the places me and my then-girlfriend (still not my wife yet...) wanted to live.

Out of all the schools I applied to, I was lucky enough to get interviews at two - University of Toronto and UNC. I remember being a bit disappointed that I didn't get more interviews, but also grateful that I had gotten any at all. I was also excited about both places - I had actually lived in Carrboro (the town next to UNC) during the last year of my PhD because my then-girlfriend-now-wife was doing her master's there, and we had really liked it. Being a deep learning researcher, I was obviously very familiar with U of T and the many relevant faculty there. I had heard a lot of great things about U of T firsthand because Alec, who has been one of my best friends since high school, was a professor there and would often tell me about its various unique benefits (which I'll get to later).

I interviewed at UNC in mid-February and at U of T at the end of March. Around the end of April, U of T rejected me. I was devastated. I had convinced myself that U of T would be an excellent place to work, and it was hard to avoid thinking that this meant UNC would reject me too. It was also frustrating to not know why - of course, you can come up with all kinds of reasons to guess, but you never get feedback on your applications and so much of the process is opaque. I remember going to ICLR in early May to present some work but staying in my hotel room for most of the conference to avoid seeing people.

Meanwhile, UNC kept pinging me to ask me what my status was (which, if you're not aware, is what a school does to their “backup” candidates while they wait for their first or second choice to make up their mind). Then, right at the end of ICLR, UNC told me they wanted to make me an offer. I flew directly to North Carolina after ICLR and we decided pretty shortly thereafter that we'd move there after one more year in the Bay Area (during which I continued to work as a research scientist at Brain).

At UNC

On the whole, I had a great experience at UNC. I actually already wrote a blog post about my first two years of professorship, so I won't say much more about it here. As we had anticipated, we were really happy we moved back to Carrboro - it's a small town where you can walk to everything, but there's enough going on that it doesn't feel boring. Then, in early 2022, we had a series of awful things happen to us. I won't get into a lot of detail here, but I'll just say that they were totally random (i.e. not really a function of the place we lived, and certainly having nothing to do with UNC) and made us feel unsafe where we were living. We thought about moving away from Carrboro, but ended up deciding to move across town to give it another try. Then, in the fall of 2022, we had a few more scary things happen and decided we should move away. It was just barely not too late to go on the faculty job market again, so I got started on my applications right away.

The second round

I honestly didn't know how things would go on my second round. While my first search ultimately had a positive outcome, the experience of being rejected by so many schools (and, especially, being rejected by a school I actually interviewed at and was excited about) really stuck with me. Imposter syndrome is a pernicious problem that doesn't necessarily require a basis in reality, but these rejections definitely provided a strong basis for whatever imposter feelings came up. You also don't get much feedback on how you're doing as a professor, and the feedback you do get is often negative (e.g. getting grant proposals rejected).

On the other hand, I had some sense that things were going pretty well - I had managed to get some funding, hire good students, and publish work I was really proud of. Some of the work I did in my last year at Brain had also ultimately become impactful (T5 and {M,Rem,F}ixMatch and friends). I also did have some explicit positive feedback, including one of the most valuable pep talks of my life from my friend Sasha. So, I was cautiously optimistic that I'd be more successful this time around.

I ended up applying to 20 schools and got interview invitations from most of them. It wasn't physically possible to do all of the interviews (I was also teaching a class and running my lab at UNC), so I had to decline some of them. I interviewed at 12 schools and got offers at every place I interviewed, and, to my surprise, all of the offers came with tenure in one way or another.

I can't emphasize how mind-bending this experience was. As I mentioned above, professors get very little external validation, and my last major source of external validation was my challenging (but ultimately successful) first round. Part of the reason that I applied to 20 schools was that there was a large part of my brain that was still certain that my second round would go similarly to the first round. While giving my job talks during interviews, I had to fight off that same part of my brain, which was silently screaming “What the f*** am I doing here?!”. But, I managed to keep my cool, and the interviews all went reasonably well.

I've spent a lot of time thinking about what was different between the first and second rounds. The most obvious thing is that I now have a good amount of experience being a professor under my belt, so I'm much more of a “known quantity”. This experience has meaningful consequences too — for example, I'm reasonably confident that I'm better at formulating a coherent and convincing long-term research vision now (because this is something you have to do constantly as a professor, but not as an industrial scientist or PhD student). Experience aside, there is a very real insularity in the faculty recruitment process, and I think the fact that I was applying from industry during the first round worked against me in certain ways. I also think I benefitted from my “early” work on large language models seeming prescient given that I submitted my applications mere weeks after ChatGPT shook things up. It probably further helped that one of the main tenets of my research program is to ensure that academia and the research community at large remains relevant in our “era of scale”. Finally, as much as I wish this wasn't the case, I know that my privileged identity (a white cis male, i.e. the dominant identity in my field) helped me in a problematic and backwards way.

The decision

I feel incredibly lucky that my second round went as well as it did. In particular, I feel lucky to have had many options that I considered equivalently good in terms of my lab's probability of success. Making the decision between these options was a challenging but good-to-have problem.

Ultimately, we decided to move to Toronto, where I'm now an associate professor at the University of Toronto and an associate research director at the Vector Institute. U of T was certainly within the “equivalently good” group, but it also was an outlier within that group - i.e., there are many factors that make it unique when compared to the other schools I considered. Here's a non-exhaustive list of these factors:

• The funding model in Canada is quite different than in the US. Most funding comes from either NSERC (which is the Canadian analog of the NSF) or industry. For faculty (like myself) who are not interested in pursuing military funding, the absence of military funding is arguably a positive. The government also goes out of its way to encourage industry grants by matching them (in some cases up to 3x). On the whole, professors here estimate that they spend dramatically less time (I've heard 10x less) working on grants than professors in the US.
• The cost of funding a PhD student is significantly lower - about 1/3 of the cost. This comes from various factors including governmental subsidies, lower overheads, not having to pay students' tuition, etc. U of T allows professors to choose their students' stipend, and so I'm able to pay my students a stipend that is more than most schools in the US, but funding one of my PhD students still costs about 1/2 as much as it did at UNC.
• The Vector Institute provides incredible benefits for faculty members, including a formidable compute cluster, engineering support, salary top-ups for faculty and students, a wonderful physical lab space, and more. It also provides less tangible benefits, like a sense of community among ML and ML-adjacent researchers.
• Perhaps this goes without saying, but U of T is incredibly strong in ML. When I count the number of professors who do work that I admire and/or want to collaborate with, U of T probably has the most.
• It's much easier to form collaborations between my lab and companies that I consult for — as long as I don't advise a student who I also manage at a company, most other things are on the table.

Frankly, I probably could have come up with a totally different list of “unique benefits” for schools within the equivalently good group. So, the ultimate decision among these schools really came down to non-academic factors. Ultimately, my family decided we wanted to try living outside of the US, and more specifically in Toronto. There are also a bunch of factors that make Toronto especially attractive:

• Many of the neighborhoods of (Old) Toronto are very livable (albeit expensive). Such neighborhoods have regions of dense housing interspersed with commercial streets. There's also a good amount of public transportation and bike infrastructure. Here is a nice video describing one such neighborhood.
• In the US, many parents spend a lot of time strategizing about schools - choosing neighborhoods to live in based on the “quality” of the school district, deciding whether they want to (and can afford to) send their kids to private school, etc. This kind of strategizing doesn't really seem to happen in Toronto - most people I spoke with just send their kids to the local public school and are very happy with it.
• Toronto is an incredibly diverse city, with about half of the population being immigrants. Apart from the obvious benefits of having lots of different perspectives and backgrounds, it also means there's an incredible selection of places to eat. Toronto and Canada are also often listed among the most LGBTQ+-friendly cities/countries in the world.
• Toronto has a lower crime rate than many large American cities. Canada as a whole has fewer guns per capita (though still arguably way too many) and dramatically fewer school shootings.
• Politically, things in Toronto (and, more broadly, Canada) just seem a notch or two more sane than they (increasingly) are in the US. The eventual possibility of getting citizenship and giving our kid(s) a non-US option also weighed in heavily.

Credit again goes to Alec and his wife Annie for educating us on all of this as we made our decision.

There were a lot of logistical challenges getting here because it's much more complicated to move internationally, not only for me but also for my students. But after living in Toronto for a few months, I'm happy we decided to move here. Academically it all clicked when one of my students was raving to me about how easy it was for him to find people to consult and collaborate with at U of T and Vector. The path here has been strange, long, and winding, but I'm happy we made it.