5 Steps to 10x your Machine Learning Productivity

A few years into my research career, I had many things to do but had limited time: Paper reading, coding, writing, managing, etc.

A few years into my research career, I had many things to do but had limited time: Paper reading, coding, writing, managing, etc.

As anyone in my situation would, I pursued the most obvious solution — I read way too many productivity books. Most of these books had a lot of the same stuff repeating for 300 odd pages. To make my life easier, I adapted the useful nuggets from these books into a five step framework. This 10x-ed my productivity.

Here it is for you to leverage in your own work:

Step 1: Rule of Three

Ask 10 people how they’d handle multiple projects & deadlines and 9 of them will tell you to multitask.

This is not the solution.

Instead, each day, write down the top 3 things you want to get done. These could be things that are time-sensitive or maybe are difficult problems you need to solve. Our brains aren’t great for storing information. They are great at thinking and problem solving.

Multi-tasking forces the brain to juggle and remember tasks and where they are in the execution cycle.

Writing 3 things down unburdens your brain and allows it to focus on what it’s great at — problem solving. Additionally, you get the nice dopamine hit when you finish and check off an item from your list. Any item that isn’t fully completed at the end of the day gets rolled over to the next day’s list.

If you do this, here are the results you can expect to happen:

Deep work:

Uninterrupted spells of flow state allowing you to do your best work on a task.

Less context switching:

No more switching between tasks when you’re in the zone. Additionally, no more background threads about the other task(s) you left unfinished while you’re working on the one at hand.

No procrasti-planning:

Procrasti-planning is the constant mental gymnastics your brain does to organize the order in which things need to be done. Eventually, nothing gets done because you spent all your time and energy on planning and not on doing. Simply put, it’s procrastination masquerading as planning.

My top 3 list for a given day

Step 2: Bitter Pilling

Once you’ve got your 3 things, here’s the next problem you’ll likely face. We all have tasks on our plates that are boring, monotonous or painful to finish. For me, these are data annotation, cleaning, writing automation scripts and reports. Your instinct might be to finish off the tasks you enjoy first and then come to these painful blisters last.

Don’t put the most boring thing off for last.

Get it done first. Swallow the bitter pill. Why? If you do it first, you can enjoy the rest of the tasks in peace. Furthermore, you’ll be happy to get this work done soon so that you can devote your time to intellectually stimulating work. Another pitfall you’ll avoid is that you won’t repeatedly postpone doing them until your plate is filled ONLY with these kinds of tasks.

Step 3: Batching a.k.a Pipelining

After you’ve accomplished Steps 1 and 2, now it’s time to pipeline. If you work with deep learning models, you’ll be familiar with the amount of time it takes to train and fine-tune them. What I’ve found is that these long windows of waiting are ripe for pipelining.

While you have experiments that you are waiting on, do something else in parallel.

For example, I read papers, answer emails or go get my workout done while my model trains.

Models aren’t the only ones that work well with batches. You do too.

Step 4: Owl or Lark?

At this point, you’re now ready to tackle peaks & troughs. Each of us have unique windows in a day when we are at our maximum effectiveness (peaks) and minimum effectiveness (troughs). The trick is to schedule our work so that we prioritize high impact work in these optimal windows.

Here’s what to do:

• Do your high impact work like coding, brainstorming, researching when you are at your peak effectiveness

• Do low impact work like automation, answering emails, logging bug fixes, etc. outside of peak hours

Owls are those who are most effective at night. They thrive when their high impact work is done in the night. Larks are those who are most effective during the day. The converse is true for them.

Are you an Owl or a Lark? Find out and maximize your effectiveness.

Do your best work at your peak hours

Step 5: Build a bridge

That’s it! Now it’s time to build a bridge for the next day. A simple mistake I used to make was to just drop what I was doing at the end of a day and come back the next day to resume. A side-effect of this was that it took me a good half hour to identify what I was doing the previous day and what I had to do next.

So:

• Find a good stopping point for unfinished work: Make it easy to pick up the next day

• Write a note to yourself on what you need to look at next. That way, you don’t start the next day with a blank slate

• Celebrate your accomplishments before you close

Overall, The reason this framework works so well is because it’s repeatable and your work compounds with less friction and mental burden. So, don’t fall into the trap of being a multi-tasking juggler.

Do this instead.

Recap: The 5-Step Framework to be a productive practitioner:

• Rule of 3

• Swallow the Bitter Pill

• Batch / Pipeline

• Find if you are an Owl or Lark

• Build a Bridge

I hope you find this useful in your own work.

🤖💪 Want more ideas to be a productive ML practitioner?

Each week, I send out a newsletter with practical tips and resources to level up as a machine learning practitioner. Join here for free →