What I learned during my PhD

The following is a collection of things I learned about academia and work during the time of my PhD.

Some of them are things I did which worked great and some of them are things I didn’t do, or did too late, and I wish I had started sooner. Apart from the technical tips, these thoughts are applicable to all fields.

Let’s jump right in.

Keep organized

You will often have multiple tasks or projects you are working on concurrently. For this, it helps when you are working with a software like Asana or Trello which helps you keep track of where you currently stand in your different projects.

Additionally, keeping a diary or a journal enables you to understand why you made certain decisions (e.g., design decisions for your method) even years after you made them. A great app for this is Logseq. Your journal will come in handy when writing a paper or a report on your work.

When meeting with your supervisor, or even when meeting with colleagues, it pays off to know the agenda or purpose of the meeting in advance. If the meeting is mainly about you communicating something to a group, slides help guide your presentation and let you supply visual components such as plots or images.

Trust yourself

This is an essential lesson that unlocks many other beneficial things for you and your academic career. Once you have begun trusting yourself and your work, you will gather the courage to publish some of your research. The work you publish doesn’t have to be a paper, it could also be a blog post - the important part is getting your work out there and available to other people. Their feedback will give you a push, and publishing something makes you an active participant, rather than an audience member watching the world of academia from the sidelines.

As for what you should publish, I want to make a bold claim here: Ideas are more important than sheer numbers. What I mean in specific is that your new system or method doesn’t have to necessarily beat the state of the art on your chosen task. Your work may be faster to compute than other approaches, or more efficient with training data, or it just offers a fresh perspective. More often than not, what is novel about your work cannot be boiled down to a single number.

Finally, I want you to remember that at the end of your PhD journey, you will be an expert. Depending on what you have worked on, your field of expertise might be pretty narrow, but within it, you will be able to count yourself among those who know the most about it.

Connect with peers

Discussing technical details of your work with a colleague can often help more than you think. It lets you see your problem from the outside and it forces you to put it into terms that someone not already involved in it understands. Also, there may be some overlap in the problems your colleague and you are trying to solve, even if you are working on very different topics.

In a similar vein, forming smaller interest groups within your institute allows you to connect with people who face the same hurdles and challenges as you, and it can have a compounding effect on your work.

Immersing yourself in your research topic can be quite fun: you can listen to podcasts, watch videos, read blog posts and papers. All of this will spark your imagination and help you explore the space of ideas around your topic. It also lets you feel connected not just to your immediate colleagues, but to the field as a whole.

Technical tips

These, I will deliver in the form of bullet points because they are unrelated to each other.

• When working on a new method, first implement a baseline in the form of the most simple method you can imagine (e.g., when working on time series forecasting, your baseline would be to predict the last known value for all future time steps). This helps you establish a pipeline and lets you fix all of the problems not related to the actual method, first.
• If you are working with neural networks, it pays off to know the details of the backpropagation algorithm (see this excellent video by Andrej Karpathy).
• Track your ML experiments, e.g., with a service such as Weights & Biases (wandb.ai) or open software such as MLflow (mlflow.org). You will have an easier time exploring the design space of what you are working on.

The most important aspect: enjoy the process. Earning a PhD can be hard, but it can also be very rewarding. Also, it’s not your fault if you are stressed - this happens to lots of PhD students. I hope these tips can help you.