Thoughts on Academia and Industry in Machine Learning Research

A recent conversation with Jay Shah on his podcast made me think more about career choices and the question of “academia vs. industry” after completing a PhD. Since finishing my PhD, I also had this conversation with many other researchers — and before finishing my PhD I asked recent graduates about this myself. So, in this article, I want to share some of my thoughts.


By construction, a PhD has a clear end. Depending on the program, country and field, a PhD is supposed to be done within 3-6 years when it is usually awarded after an official defense of the research work. This is in contrast to most other careers and jobs, especially in industry but also in the public sector. Even though a PhD is often considered as a qualification for independent research and thereby acts as the entry to an academic career, it is commonly assumed that most PhD graduates do not continue in academia. This also matches my impression and surveys among PhD students in the Max Planck society.

As a result, the discussion around academia vs. industry naturally comes up towards the end of a PhD. There is a lot of discussion of this topic on the internet: For example, there is Elizabeth Bondi-Kelly's blog on the academic job hunt, Nathan Lambert's article on the industrial job hunt, Arne Köhn's thoughts on deciding between academia and industry, or Russ Salakhutdinov's thoughts on staying in academia — all of which I enjoyed reading. The debate is also happening on social media; on Twitter, for example, there are communities trying to increase visibility of moving from a PhD into industry, not only in machine learning.

In this article, I want to share my thoughts on this topic after finishing my PhD in 2022 and then moving to industry. Of course, this is an extremely subjective opinion on this topic that has been shaped significantly by my personal experience — which has predominantly been positive in both academia and industry. Nevertheless, I want to take a look at the respective systems rather than individual jobs. So, instead of focusing on what a post-doc looks like compared to a mid-level (machine learning) engineer in industry, I want to highlight that academia and industry play according to different rules, favor different activities and provide different perks.

Different systems, their rules, activities and perks

Most career trajectories follow specific rules that are defined either explicitly (for example, through regulation) or implicitly (through hiring practices, culture, etc.) by people following the same career path. In academia, expectations are defined by other academics, universities, organizations providing funding and so on; in industry, these rules are set by other professionals and companies. In general, I believe that everyone wanting to pursue a specific career has to be aware of these rules. I am not saying that it is not possible to transition between academia and industry, but I found it quite useful to realize these different expectations. For my German readers, this thinking in terms of system rules is very much inspired by the career advice of Heiko Mell in the German "VDI Nachrichten", a magazine from the German association of engineers.

There are plenty of examples for these rules. Starting with the CV (or "resume" in industry). In academia, one usually does not list a lot of bullet points below each position to describe the completed work. Instead, publications, grants, reviewing, workshop organization, etc. are all listed in separate categories. Most resumes in industry do not even include many of these categories. Instead, prospective employers are more interested in the actual past work — used technology, business impact, leadership experience, etc. Similarly, interview styles usually differ: research talks in academia vs. coding and culture interviews in industry.

After securing a job, the actual contracts also differ considerably. For example, post-docs are often employed using short-term contracts tied to a funding source or grant and regulations are often similar to public service (at least in Germany). In industry, instead, getting an actual salary (vs. a scholarship) is the default and regulation in terms of short-term contracts, notice periods can be very different from academia. This also impacts quitting a job. In academia, switching institutes after a PhD and/or post-doc is often recommended. Nevertheless, collaborations with previous institutes might continue. As a result, sharing knowledge and even resources across institutes is quite common. In industry, employees will often lose access to their account and company hardware as soon as their employment is terminated (as demonstrated by the recent tech layoffs).

There are also differences in terms of roles and activities. For example, while most professors are also teachers, advisors and managers, these roles are more separated in industry. Often, there are separate career tracks for engineers and engineering managers and most engineers do not have an obligation to teach X hours each semester. There is also less publishing in industry. This means that academics will spend a significant portion of time writing, revising or reviewing papers. In industry, this is somehwat replaced by writing (internal) design docs, RFCs (requests for comment) or patent applications. I believe that maximizing the time spend on interesting or enjoyable activities (be it coding, writing, organizing, etc.) should be an important consideration when deciding between academia and industry. Thus, these differences can have very direct impact on job satisfaction.

Finally, and most obviously, perks are very different. In discussions, I find this is often summarized informally as "freedom vs. money". However, it also involves contract length/conditions, travel stipends, home office, etc. Depending on people's background and culture there might be positive or negative associations with these respective perks. So, I decided to devote the main portion of this article to these differences.

How free is academic research really?

The high-level promise of a long-term career in academia is research freedom — of course not immediately as PhD student or PostDoc, but maybe as tenured professor. In my view, there is some truth to that as tenure often provides high job security and some degree of unconditional research funding. In order to get tenure, however, researchers need to be "successful". Usually this means publishing in top-tier venues, getting grants, giving talks, contributing to the research community etc. This will be more difficult when working on topics that the research community, funding organizations or universities deem unimportant or less interesting. This means that there are strong incentives to align research topics with what most other researchers find important. So research freedom is somewhat relative to current trends, even though one can decide how much to align with the community. The same also holds for how to organize research. There might be some freedom in how to organize research, but when requiring specific resources or support there will usually be plenty of constraints and rules to follow.

Even if there is considerable research freedom, this also comes with responsibility. In my case, this had a quite profound impact on how I identify with my work. The disadvantage of freedom is that deciding on research topics, in particular early in a research career, can be incredibly difficult. With the freedom to choose projects also comes the difficulty of justifying this choice when projects go wrong. In the end, there is nobody else to blame. It is easy to associate oneself with these research outcomes. This is somewhat reinforced by the academic culture, implying that one is always working for oneself. I spent so many weekends working for my research, my reputation, etc. Of course, on the positive side, success would also be associated very much with myself — including my degrees, my funding, my awards. Nevertheless, I believe this is a key reason why PhD students and young researchers are prone to mental health issues because it is incredibly difficult to unlearn this thinking.

Overall, I learned that academic freedom can be a double-edged sort. With fixed-term contracts and pressure to align with the community, I feel that freedom is limited before obtaining tenure. Even with tenure, I think this freedom also entails the personal attachment with both success and failures which can be hard to handle — although I cannot judge this personally.

How capitalistic is industry actually?

When moving from academia to industry, it is often said that this academic freedom is "sold" for money. Looking at salaries in industry, there might be some truth to it. However, in my opinion, this argument ignores the fact that pay and work conditions in academia can be very bad, skewing the statistics. Anyway, besides this pay gap, I think the saying tries to emphasize something else: the transition from an academic system into a capitalistic one. Of course, academia also runs on money, but the impact of the micro or macro economic situation is less direct in academia than in industry. Often, this saying also has negative associations within academia. However, I do not think that these negative associations are always justified.

Clearly, companies can exercise much more control over employees than universities or research institutes in terms of what they work on and how work is organized. In many countries this is the very definition of tenure — high job security allowing to study uncomfortable research questions despite the current economic and political siutation. In industry, in contrast, projects may be discontinued or started more or less at will and there is generally less knowledge transfer with the "outside world". Re-organizations are much more common, meaning it can easily happen that roles get changed, moved across teams, or even eliminated (not always causing a layoff). Hiring is also much more dependent on the economic situation compared to academia, meaning that ones salary depends on the current hiring market. All of this can cause much higher personal uncertainty because of less control over the overall situation. This is often perceived as negative but might just mirror a key element in capitalism — higher salaries come from taking higher uncertainty. However, this also means that perks and salaries are much more negotiable. Of course, early career academics usually also face high uncertainty in terms of short-term contracts and securing founding. However, I think that many academics do not perceive it this way, possibly because securing a post-doc or funding seems much more controllable individually.

In the end, I learned that there can be higher (or at least different) uncertainty and more constraints when working in industry. The former is often related to giving up control and I found the latter rather freeing at times, especially when I still have a lot of liberty in how I solve problems.

My Personal Decision-Making Process

Considering all of the above, everyone will have different preferences and, often more importantly, different constraints. Personally, I decided to go with industry for four key reasons:

  • Skipping the post-doc phase: Personally, a post-doc felt rather unattractive. The expectation is to move to a new institution, ideally a different country, after the PhD. It is very common to do a post-doc in the US, for example, and then move back to Germany for a group leader/tenure-track position. Moreover, post-docs usually have rather short time frames. This makes it hard to pursue longer-term research agendas or align an academic career with personal goals (a partner, family, other responsibilities, etc.). I felt that industry is more flexible in some of these aspects and also pays better.
  • More focus on engineering: I always loved the engineering part of my research. Most of my work is open-sourced and many of my papers are cited mainly because of that. However, I felt that good engineering practices are less appreciated in academia compared to industry. I still like writing papers and giving talks, but I also want to see my coding contributions being appreciated.
  • The option to go back to academia: Throughout the last few years, I saw more and more researchers and engineers switching back from industry to academia. While these cases are probably the exception rather than the rule, it showed me that this is not impossible. Thus, I do not feel like I am closing the door to this career path entirely.
  • The people I work with: Finally, I did an internship and very much enjoyed working with my team. I felt these are people I can learn from and grow with. This was probably a very important reason for me to join Google DeepMind and I met many other young researchers that preferred academia due to negative experiences during internships or because of specifics academic they like to work with.


In the end, deciding between academia and industry is a very personal and difficult decision. It has many implications in terms of the daily work, the perks, career paths or the compatibility of work and family. I am convinced that understanding the respective systems and their incentives can be extremely useful in making this decision. However, the respective rules and expectations are evolving and rarely written down. That's why I had to talk to tons of people in academia and industry to get a rough picture to help me make a decision. I hope that this article contributes to documenting some of these differences and their effects.

What is your opinion on this article? Let me know your thoughts on Twitter @davidstutz92 or LinkedIn in/davidstutz92.