Reflection of My Mid-PhD Journey
Milestone: I had just passed my candidacy exam last month. It’s probably a good time to document some self-reflections after going through the ups and downs in this two-year doctorate journey.
🔥 The Ups
🦇 Advisor-Student Relationship
I am fortunate to have Steve as my advisor. I particularly appreciate his (1) weekly one-on-one meetings on top of lab group meeting, (2) occasional casual coffee chats to bounce off research ideas and opinions about the state of AI, (3) being supportive of students’ exploration of research interests while giving both high-level and low-level, technical and general, feedback, and (4) being flexible, responsive, and trying his best to alleviate the stress off his students.
Knowing I have his full support enables me to take ownership of my research career. While Steve has not worked on multilingual projects before, it makes me feel priviledged and grateful that he is open and supportive of my research interests in the coming years of PhD. In fact, through him, I was introduced to the BigScience community and met many fantastic and passionate collaborators, which eventually led to some projects/initiatives I’m proud of, such as BLOOM+1 and starting a grassroot research group for Southeast Asian languages.
🌸 Collaborative Research
I am a huge fan of open-source collaborative research. To me, the upsides significantly outweigh the downsides. Collaborative work invites diversity of perspectives and communication of findings across multiple fields, which often generates exciting project ideas and keeps me abreast of the latest NLP development.
Scaling from a two-person project to 10+ people project required me to develop a different type of project leadership, especially on how to best manage resources, such as allocating human/compute resources and large-scale experimental tracking. Most importantly, I’ve improved my communication skills and overcome Imposter’s syndrome as I had to listen to various feedback and ultimately decide on the project direction.
🐻 Doing Research at Brown
One word I’d use to describe the research culture here at Brown is chill. I don’t feel the external pressure to prioritize the quantity of publications. Such a non-competitive research environment offers me the opportunity to engage with research in a healthier manner, where I get involved in different projects because they are intellectually stimulating (internal motivation) instead of the constant “publish or perish” reminder (external motivation). It also allows me grow in different dimensions instead of solely optimizing the publication metrics.
Brown also promotes a collaborative research culture. For instance, we have a Superlab collaborative research group involving Steve’s, Ellie’s and Chen’s students, where we meet every two weeks to get feedback on projects or conference abstract submissions. Another notable group is BigAI for robotics research endeavors.
💡 Things I’ve Learned
😩 Stress.
I enjoy working on multiple things simultaneously, as I am intellectually stimulated by diverse topics and potential opportunities. Therefore, I often run into the issue of taking up too many responsibilities. In fact, I experienced burntout in my first year of PhD, and it took me several months to recover my level of productivity.
Since then, I’ve learned to say no and become more self-aware of my manageable stress level, including noticing signs of imminent burnout (e.g., self-sabotaging behaviors or a reinforcing feedback loop of procrastination). I’ve also become more aware of the antagonistic combination of projects; for instance, no two engineering-heavy first-authored projects at the same time.
Aside from self-awareness, it helps me a lot to have close and supportive friends as well as develop simple self-care routines1 when there are a lot of deadlines. I also find it necessary for me to have passions outside of work — activities like social dancing2 would instantly lift my mood during the low days.
🛑 Project termination.
I’ve worked on a project for more than a year (to be exact, 16 months) that I had to move on because our efforts were futile despite how straightforward and seemingly feasible the idea was. It was a difficult decision because of sunk cost and the possibility of it working out if we stuck to the problem longer. In the end, we exhausted all possible hypotheses about the obstacle and pulled the plug on the project.
Project success being a non-linear factor has made me rethink the right attitude to approach research. My aforementioned experience has taught me how to properly design simple experiments to quickly confirm the feasibility of an idea and how to set up good baselines. Instead of being overzealous on jumping into a problem, I now spend more time sharpening the problem and understanding the tension — simply put, contradictory theoretical or empirical findings on a problem — such that I can properly test the hypothesis and derive interesting insights from either positive or negative experimental outcomes.
👅 Developing Research Intuition / Taste
Two years ago, my opinions were easily swayed by superficial attributes such as publication venues and the source’s authority. Now, I am more capable of evaluating papers or Tweets with a more critical mind. I have developed a better intuition and sense on what research endeavors would significantly impact the NLP landscape.
Before, every paper was equally cool; now, some papers should be prioritized by reading more deeply and following up with future work.
This comes from talking to many researchers and practitioners, tracking how certain vocal researchers’ opinions pan out after a period of time, and keeping an open mind.
While reading a lot of papers helps, I find it more productive to analyze the opinions of experienced researchers on them. Why are the papers useful? Are they rehashing what we already know, or are they finding something novel? It is sometimes still hard to disentangle popularity and hype from the papers, but I’d like to think I have gotten better at it.
Remarks
I came straight into PhD from undergrad. To some extent, I went against the common advice that I should go into PhD only when I really wanted it. At that time, I wasn’t 100% confident if PhD was the right decision at that time, but I trusted my instinct that I enjoy research+engineering far more than sole engineering work (given my diverse work experience in undergrad) and that working with my advisor Steve would be enjoyable (after talking to him and his students). I can now confidently say that I’ve made the right choice. Excited to see how I grow in the next three years!