My expectation was that, after the doctoral study, I would be a person KNOWING A LOT! But, more of being an intellectual who knows what he doesn’t know and how to study the UNKNOWN.
It turned out … I had a nearly impossible goal. No one would be able to tell exactly how to study the UNKNOWN without knowing it already. The more I studied, the more I thought I barely knew anything. The vastness of uncovered truths, principles and knowledge was far more profound than I could’ve ever imagined.
Personally, starting a program in early 30s, though not being too old, may have triggered a few mental crises.
In particular, the first few semesters were the most challenging. Being an industry professional for a few years, I didn’t really think of myself of anything other than an SME (subject matter expert).
I never took a seat of teacher, expert, or researcher. So, when I was confronted by the idea of “producing a novel piece of work”, I said to myself, “Uh, who am I to create knowledge????”
I was probably a reasonable person in terms of self-awareness. I wasn’t a superstar among those famous professors. I was there to obtain new knowledge. Meanwhile, internally, something was bugging me.
If I were to try to contribute something novel, the most logical place I could start would be improving the existing knowledge. I had a constant fear of “maybe I am improving the wrong thing.” It could be that what I was trying to improve was never useful, or we could never catch up to a level that my novel results would matter. In short, I was afraid that I might have been wasting time producing meaningless results.
Let me be more concrete. I was doing a PhD in areas of ML and Optimization. The topic of my thesis was about designing optimization algorithms for machine learning. At that time, deep learning was trending. I think most of people of my year were doing something of deep learning. I had to bet on what I believed the useful direction would be.
Designing optimization algorithms for deep learning is the mainstream. The most famous and widely used is SGD and ADAM. My goal was to improve some of its inefficiency! But, would my improvement be meaningful and impactful?
Remember I was doing this from 2016 to 2021. I could either be wrong or right. Let’s use the AI boom to justify my direction. The optimization algorithm is not the bottleneck of AI development, but data, architecture, and even computation. My efforts of improving the algorithms had a marginal effect. If that was true, I was indeed wasting a lot of time on something that may not even yield any impact.
From a personal perspective, my journey was started with a crisis. After the crisis, I came to understand that science is only meaningful collectively. There is a beauty in it.
In retrospective, I am a very small piece of the scientific community. I wasn’t achieving anything that’s remotely close to the scale of “paradigm shift”. Rather, I was probably just being a witness of a tide of scientific revolutions. Maybe, like all other Ph.D.s, I was just a foot solder in testing and verifying some ideas.
For anybody who is about to embark on the journey, I wish you to understand the mental challenge pretty well. Specifically, what I am trying to get at is that, there are “unknown unknowns”. There are plenty of things that you don’t know that you don’t know! And research projects are famous for finding an unknown that you didn’t know.
The mental part is to be persistent for long enough, and patient for long enough. You’ll be creating something that wasn’t seen in this world.
My education background prior to Ph.D. was an M.S. in Industrial Engineering, mostly focused on the applied side of Operations Research. After the master program, I started working in the tech industry. It was around my 2nd year of being a data scientist when I started thinking of a Ph.D. seriously.
By the time I came up with this crazy idea, I’ve already worked on solving some of the most complex problems in the field. I was amazed by the intricacy of the natural beauty of problem solving. Meanwhile, I was also deeply puzzled by and confused about the complexity of the problem solving techniques in general.
Then, Google DeepMind released AlphaGo and beat Lee Sedol in March of 2016!
It really caught my eyes, and I thought Reinforcement Learning could be THE ultimate source of power in solving some of the most complex problems!
Apparently, I didn’t land in the field of RL. Like I said, it was quite hard to navigate through a Ph.D. journey without some surprises and uncertainties. To make my story more dramatic, I quitted working with my first advisor and went with another IN THE END OF MY 3RD YEAR!
I will talk about some of those necessary struggles in other parts.
Back to the story…
The idea of doing a Ph.D. kept me up at night that I eventually stopped overthinking and pushed myself one step further: to figure out HOW!
I was fortunate enough to keep in touch with my advisor from the master program at Lehigh. The professor educated me a lot on what I would need to make it happen. Among many things, I knew securing the funding was the most important part of it.
It was generally true that any top level professors would not like the idea of funding a part-time Ph.D. student. With many not-necessarily-more-capable-but-definitely-having-more-time applicants applying their apprenticeship every year from all over the world, the incentives to spend their funding on someone like us was fairly low.
Taking my department at Lehigh as an example. When they considered my application, they knew I would be the first part-time person for many decades in the history of the department. If I were the admission committee, I would worry if this guy could ever be able to finish the program.
The story of me getting into the program was not dramatic. My employer agreed to cover 100% of tuition and other expenses … while continued to pay me as a full-time data scientist. I got the chair and my advisor from the master program rooting for me, and others were somehow convinced. That’s about how I got into the program. Honestly, I think having my own funding and a familiar face were probably why I got accepted.