This post is a reflection of my experience during my PhD and summarizes what I think were important in helping me succeed.
1. Physical and mental health
I have been suffering chronic pain in my joints and muscles for a number of years now. There were days when the pain was so agonizing that I pitied myself for this misery at such a young age. But having endured the hardship only fortifies my zeal for life. I must maintain a strong health, physically as well as mentally, to be able to do all the things that I want to do.
A few simple things that I did and will keep doing:
– Exercise fifteen to twenty minutes a day
– Walk to work (this usually took fifteen minutes from my home to NICTA)
– Drink plenty of fluids, mainly water but sometimes cow milk, especially before I go to bed and as soon as I wake up in the morning, and also after around one hour
– Take a break from the monitor every now and then
– Have lunch with friends away from my desk.
This has been a very positive change for many of us as previously we just ate by ourselves at our desk while staring at the computer at the same time
– Eat mainly healthy food.
I don’t force myself to abstain from what conscientious people would deem bad or junk food — sometimes I eat them but I refrain to a very moderate degree
– Enjoy weekends with my family and friends.
It’s hard to quantify the impact of this but coming back fresh and fully recharged after a rest over the weekends made me quite productive.
2. Talk to my supervisor regularly
Ideally once every two or three days. However most supervisors are very busy so consider it good good enough if they find enough time to chat weekly. This is absolutely important during the early period (first year) when students often struggle and need help the most. After the first year, most students become more independent and would actually enjoy more time and flexibility to play with their ideas, the frequency of the meeting can be reduced.
3. Treat it like a job
That means going to work every weekday, from 9 – 5. Sure there were days before a deadline when I needed to stay late, or there were a few consecutive weekends that I had to forfeit. But in general I tried to stick to the working hours of a regular job. This also allowed me to keep a work-life balance, most importantly to have time for my family and friends.
4. Run experiments over nights and weekends
It is the nature of machine learning research to run many experiments on a computer. One reason for this is that we often have to try many different configurations of the model parameters before we get a good setting. Whenever I have a lot of experiments to run, I would schedule as many experiments as I can and let them be run over nights or weekends. When I came back the next mornings, the outputs were generated and ready to be analyzed. This was easily one of the most effective strategy.
5. Own my research
I figured out very early that I must take charge of my research. It is up to me to acquire the necessary background knowledge, to review the literature, to figure out the problems that I want to solve and how would I go about solving them. Of course Alice and Edwin were always very open to new ideas and discussions or making suggestions, but the rest is my responsibility. Realizing this gave me the freedom to explore paths that had not been taken before without having to wait for anyone’s approval.
6. Get my hands dirty
I produced three high quality papers in the span of about 7 months, a record that I am very proud of. I believe the main factor was my willingness (often eagerness) to get my hands dirty. When I came up with a good idea, I would code up the solution in a few weeks (typically after deriving the underlying mathematical operations). My goal was to have a working version very quickly and then later optimize for speed or performance. It’s hard to guess whether or not a model is suitable for a given task and the associated dataset, so having a decent working model is far better than polishing an imaginary model.
7. Time management
I applied some time management techniques. One of them was to have a todo list for each month, week, and day, and to prioritize on the importance of the tasks. For example, the most important task was often also the most challenging task, so I would spend the morning when I was quite productive to tackle these problems. I limited email checking to a few times a day (although some time I was guilty of perhaps checking it more than needed).
8. Project management
As a student with scholarship support, I wasn’t assigned any project. Most of the projects were my own research project, however I still applied some fairly basic project management principles:
– Be clear about the goal of the project (often the research objective)
– Break the project into manageable chunks and tackle the most difficult one first
– Document the progress (the model, mathematical derivations, experimental settings and results). This made writing a paper less difficult as the skeleton of a paper is often contained in the document
– Use collaborative tools (github for sharing the code, Google Drive for sharing documents and papers, etc)