from Guide to the PhD on Jan 29, 2023
What I learned in my PhD.
Reflecting on my several years in graduate school, I realized I've come a long way — in how I tackle problems, find problems, and more. Here are the top reasons why I'm glad I did the PhD, and why you don't need a PhD to learn the same.
I started the PhD more or less because I got accepted — not for any deep-seated, inspirational reason. In fact, I didn't know what the program really entailed, and I certainly wasn't sure why I was choosing to stay in school, when I in fact struggled in school.
Now that I've finished though, I'm really happy I did undertake the program. This post is primarily written for three different audiences:
- Anyone that has just started a PhD and is looking to see how to grow. In which case, this post serves as a rough guideline for how to maximize value from the program. Remember that the value of the program is inter-personal, which you can only reap by interacting with other researchers regularly.
- Anyone that doesn't want to pursue a PhD but is looking for the non-material benefits that such a program provides. Follow the tips in this post to understand what skills graduate school teaches and how you can pick up those same skills, even without undertaking the PhD program.
- Anyone on the fence about starting a PhD, unsure of whether the perceived downsides are worth the upsides. If this is you, make sure to check out Why pursue a PhD? Is it for me? and use this as a guide for the pros of a PhD program. The skills themselves are not the benefits, but an environment streamlined for building those skills is.
In this post, I'll talk about what I learned, then cover a different angle on these learnings — how you can learn the same exact skills and reap the same benefits, but without going through a PhD.
The PhD program is primarily for three main purposes, and as you may notice from the title, all of these purposes are geared towards other people you meet in the PhD.
First, the program provides "brand" guidance. In particular, you learn how to contribute to the sum total of human knowledge and how to build a brand around your contributions. For the only time in your life, you have a myriad of unique advantages that allow you to focus entirely on building this:
Committee: Several, already-famous researchers — a.k.a., your advisors — have dedicated time and resources to discussing your contributions with you. In particular, the qualification exam at Berkeley is designed for your committee to assess your thesis plan:
- Impact: Is your vision worth pursuing? If accomplished, would the impact be significant enough to warrant attention? Are the objectives too narrow?
- Time: Is this vision possible to pursue in your program's timeframe? Are the objectives too broad to pursue?
- You: Are you the right person to pursue this agenda? Would you need assistance or mentorship, and are there mentors available?
Peers: All your peers are actively undergoing the same internal struggle, and these peers are more than struggle buddies — many of them succeed in making significant research impact. This is not only exciting to watch and learn from their rise but also mind-boggling to think that such famous researchers are your friends.
- You: Your success or failure in the program is reliant almost entirely on your ability to find and define this brand, then use it as a vehicle to contribute to human expertise in some area. This pressure and simultaneous focus is unique to a graduate program. There are no distractions that matter, such as grades or promotions.
Second, you ingratiate yourself in a research community, and this community is particularly special in two ways:
- Shared interests: Critically, you network with researchers that share interests with you. This is an entire body of talented individuals that all care about the same problems you do. They speak the same language, share the same expertise, and have themselves made significant contributions.
- Established mode of networking: If you take a step back and consider: Conferences and journals are an especially effective mechanism for discovering and connecting with researchers that share interests, and few analogous mechanisms exist for interests beyond academia. How would you find other database experts? Networking nerds? Origami hobbyists? There are discord forums and social media tags, but there is no unified, standardized way of finding like-minded hobbyists.
There are three types of communities that you may encounter in your time in a PhD program:
- Within-university: Your first and foremost community of like-minded individuals will be your lab, naturally. Other related labs in your university, possibly labs in a shared consortium, are also great sources of collaborators, peers, and mentors.
- Across universities: You will over time get to meet faculty and students across universities, possibly by giving talks, collaborating, or just by chatting at academic events. It's a pleasant and memorable surprise when you already know their work, and the same is true vice versa.
- Industry: Through internships, corporate events, and workshops, many industry labs will engage academics as well. This is particularly true of the AI field but may extend to select other fields as well.
The combination of the above unique facets of a PhD make it an idyllic program for developing certain skills, which we'll now go over.
No matter the PhD program, your success is judged by your quality of research. In the AI field, this is in turn assessed by the quality and impact of your conference publications. As a result of this singular focus on your research impact, whether implicitly or explicitly, you learn two important sets of skills.
You learn how to form a vision and break down this vision into 4-6 month long projects — ie, a paper. In many cases, you also learn to do this in the reverse order: Start with a series of projects and later string a vision through. This skill allows enables you to generally do the following:
- Trains you how to set objectives for a rough timeframe. Although you may not know in advance how long a paper will take, you do learn what a reasonable scope for a paper is. This allows you to differentiate between a multi-year objective and a half-year objective, at a rough level of granularity.
- Trains you to find problems, then solutions. In general, the most effective stories in research are motivated by a well-defined problem. This is true of every famous paper: ResNet was motivated by deeper networks not improving accuracy. BERT was motivated by missed opportunities in unidirectional, left-to-right language models.
You learn how to execute a vision, breaking down grand goals into more palatable objectives. In many cases, progress on a research agenda is dependent entirely on you, so it's a sink-or-swim situation. Figure out how to execute, or the project is never completed. There are two facets to execution.
- How to execute the project yourself. This is the most effective method for learning — when you execute yourself, you simultaneously learn how best to plan. Quite simply, don't plan, and you go nowhere no matter how quickly you execute. Plan too ambitiously, and you lose motivation to execute. At some point, you find the right blend of easy-to-do but impactful experiments to execute, along the line of "Fail Fast" in What defines a "good" researcher?
- How to lead execution, when advising undergraduates or master's students. In some sense, you learn how to plan effectively based on your interactions with others. Design tasks too difficult, and no one will be able to complete them. Design useless tasks, and no one will want to complete them. Learning how to mentor effectively teaches you how to plan for others and for yourself.
You learn how to communicate a vision, especially complex ideas and even half-baked ideas. This is one of the most important lessons: You don't flesh out a story before communicating it. Instead, you communicate to flesh out a story. This is what lab meetings, informal talks, and in-person interactions with your peers are for. There are two particular skills for communication:
- How to sell your work in 30 seconds. To be clear, you don't learn how to develop the perfect pitch quietly in a corner — even after your PhD. Instead, you learn to sell your work often, updating your pitch based on feedback and reactions. This process is what you master, and it arises because of common practice in the program: All researchers will ask you, "What are you working on now?" This is more or less the academic version of your default question for friends you're catching up with, "What are you up to now?"
- How to sell yourself in 30 seconds. Over time, you'll also develop a personal pitch out of necessity. The necessity is not due to your thesis requirement but due to the sheer numbers of times you'll meet a random researcher who asks you, "What do you do?" The most common pitch is your year and a recent project: "I'm a third-year PhD student studying computer vision — specifically few-shot object detection at the moment."
All in all, you learn all the skills need to scope, execute, then pitch your work.
One of the most common questions I get is: "Did you need a PhD to learn that?" Honestly, you technically don't. These skills can be learned in any setting really. The unique part of the program was the pressure: Your success depended on nothing other than your research contributions. However, with the right discipline, you could pick up the same skills in industry.
- To practice breaking down a vision into shorter-term projects, do the reverse: Take projects — either your own or other's — and connect them under a bigger theme. The theme should be a proposition that can be argued. For example, linking together FBNetV2 and FBNetV3 under "FBNet" or "neural architecture search" is not particularly interesting or informative. Instead, we could argue for the theme "Neural architecture search can be computationally cheap and accurate." Right now, pick a few projects and find a common theme. You can substitute out projects to make the theme more cohesive. Projects don't need to be papers — they can be libraries, movies, or games.
- To practice executing a project, in particular to estimate timeframes, take your tasks at work or at school and estimate the number of hours or days until completion. Do this for several tasks, and make sure to log the actual amount of time. Over time, you should find that comparisons between your predictions and ground truth will improve your ability to estimate timeframes. Right now, pick a task and estimate its completion time in hours or days. When you complete the task, compare the actual to your predicted completion time.
- To practice communicating a vision, share what you're working on with your coworkers and friends regularly. Make sure to follow all pieces of the story: Start with the problem statement, state the intuition, then the method. Furthermore, ask your coworkers what they're working on, what challenges they're running into, and how they're dealing with those challenges. Right now, pick a 1on1 or coffee chat to discuss a project with someone. If you're a student, share your research or side project with a friend. If you're in industry, share your project. As a bonus, try to regularly meet different people across the company, so you have to regularly introduce yourself and your work.
Try these simple actions to slowly build up the relevant skills over a years-long period. In or outside of the PhD program, these skills take time to learn in either scenario.
These are aspects of a PhD program that aren't universally true, but the following were highlights for me.
First, I learned what makes me excited. This alone made my entire academic career worthwhile.
- There's an interesting phenomenon in research, of lack of enthusiasm: At a conference, talk, or even dissertation, very few graduate students are enthusiastic about their work, deep down inside. At conferences, researchers are now re-sharing for the millionth time a project that was completed 6 months prior. The same goes for dissertations, talks, and any other presentation. Instead, they're actively thinking about and working on a newer, exciting problem.
- This was true for me as well, so despite stapling together a thesis to graduate with, I found a new, shiny interest in AR/VR at the end of my PhD. Even though it took me 4 years to find my "passion," the degree of excitement I feel makes the journey worth it.
Second, I found an inner ring of welcoming teaching faculty. The vast majority of cliques in academia are exclusive, but I'm extremely fortunate to have worked with and known very welcoming faculty as well.
- Among the most inviting are the teaching faculty. This includes faculty I TA'ed with, lectured with, and those that I got to meet with. At the end of my PhD, as I had a few opportunities to lecture, I got to meet professors I'd looked up to for many many years. This was the first time I've met my inspiration and not been disappointed. These are faculty that the students universally admire, including my peers.
- On a side note: during the height of the COVID lockdowns, I also learned a really bizarre fact — academia is quite closely linked to industry and government at the highest levels. Faculty and deans were speaking government officials, technical leads for national response teams, and high-up executives at large technology companies. I had the opportunity to meet people I never would have otherwise.
For the most part, I can summarize this entire post in a few words: The benefit of the PhD is inter-personal, both in terms of the people you meet and the skills you develop. To both your advantage and disadvantage, the communication you learn is a rather specific one: You learn how to communicate with other experts in your field. However, this is a starting point. If your cutting edge research isn't easily understood by experts, it definitely won't be understood by anyone outside of your field.
Want more tips? Drop your email, and I'll keep you in the loop.