Advice for first-year students, from first-year students

Student 1

Research

• It is good to keep a weekly log (or whatever cadence works for you, I find daily logs too granular, and struggle to actually remember everything if logging more than a week's worth) of research progress.
• A log should record the things you've tried out or ideas you've had. Make sure to include both your successes and failures. The former is useful for keeping motivated, and the latter is useful in helping you avoid the same mistakes.
• Your logs may be useful when you're writing a paper (what exactly was so hard about a project?), or in helping you identify bottlenecks in your research progress (there's a todo item that's been in the log for much longer than expected, etc)

Institutional Knowledge

• Your fellow labmates (especially ones who have been around for a bit) are usually a great source of institutional knowledge!

Meetings

• It is helpful to start a meeting with a brief recap of the previous meeting's action items (see next bullet). This helps attendees reload context. The act of preparing the recap beforehand will also
• Unless the goal of a meeting is a brainstorming session (e.g., for research ideas), meetings should have an agenda and result in action items (tasks that have explicit outputs). I like to call this the “no agenda, no attenda” rule.
• Even for brainstorming meetings, it's a good idea to do some prep beforehand to guide the session (e.g., what topics will be discussed?)

Misc (fun)

• There is often free food at events (like distinguished lectures), or leftovers from lab meetings, having a spare container is a good idea, so you can save time for a meal that day or lunch the next. This is not to mean you should stuff your container with free food; please make sure that the food you are taking is actually up for grabs.

Student 2

My advice

Fail hard fail fast: Don't create the perfect infrastructure first; create a proof of concept, and only engineer it to the extent needed to make sure there are no confounds. Optimize only when proof of concept works (or shows signs of working). Prioritize iteration velocity: If you're trying to get a method to work, try it on simple environments that don't strain your compute/don't require too many resources and ablate through all parameters

Pick courses that help your research (in general) to the point you can, and then don't worry about the grades Getting a 3.7 requires 4 hours a week. Every additional point requires an extra 4 hours a week. 4.0 requires 16 hours a week. 3.8 is the point of diminishing returns. Please don't fail the course based on this advice

Tips that an older grad student gave me

Be intentional about how you spend your time

• Plan your week and prioritize
• Sandwich meetings
• Give yourself large chunks of focus time
• If you can’t focus, change your environment

Focus on the process not the result

• Don’t spread yourself too thin
• 1-2 research projects max

Get sleep, you will be more productive. Take time off, it lets your ideas marinate

Focus on your gradient, not your status

• Try to be the least experienced person in the room, it is the greatest opportunity to learn
• Is your ego getting in the way of your goals?
• A smart scientist listens to ideas from all people — newcomers to experts

Don’t ignore your mistakes, study them:

• Sometimes you run a 48-hour experiment and find that you didn't set some hyperparameter correctly, or that something worth “tracking” was not tracked and now you have to do it again
• Don't rely on the good intentions of your future self. Create a mechanism so that this mistake never happens again (for instance, use asserts)
• Maybe a mistake could have been avoided if you had just done better experiment tracking. Use good tools.
• Maybe you went on a wild-goose chase trying to implement an algorithm that has no/weak theoretical foundations — next time you should spend more time understanding the method, and what it builds upon

Networking

• It is obvious when you are only trying to “network up”
• Recognize diverse talents
• Uplifting others is not only good to do but also can help you down the road

For your first project, start by joining a project with an older grad student or team

• Learn the flow of research in your lab
• An intermediate for talking with your advisor
• Gives you time to learn about various topics before diving in

Tips/Advice from my advisor

It’s important to articulate the “what?” and the “so what?” when discussing the challenges to your research problem.