Part-Time Learning for a Full-Time Career:
Wherever you are in your career — A student, a fresh graduate just into work, or a grizzled veteran, there is one mindset that you should…
Wherever you are in your career — A student, a fresh graduate just into work, or a grizzled veteran, there is one mindset that you should never abandon. That is the growth mindset. A key component of the growth mindset is the habit of constantly learning new things.
This is incredibly relevant for machine learning practitioners.
Let’s wind the clocks back a little over a decade. There is no deep learning. Researchers focus on feature engineering and handcrafting classifiers to solve their problems. The year 2012 put paid to that practice. Since the arrival of Imagenet, the dormant field of deep learning has exploded into life, and (almost*) every paper since has proposed some neural network or related algorithm as the greatest thing since sliced bread.
To drive this rambling preamble home, think of the architecture choices we use in our machine learning problems today.
A vast majority are based on transformers which came out in 2017. Convolutional networks, while not obsolete, are almost never proposed in new research. AI can now write code, create art, understand human speech perfectly, and more. A lot of jobs that previously required humans in the loop now are fully automated.
Machine learning is moving at a tremendous pace, and it’s not slowing down anytime soon.
We need to find a way to keep up.
So, how do I constantly learn you ask?
Learning anything new is hard. While our brains enjoy the pursuit of knowledge, we experience inertia during the early stages. We then use the algorithm in our head that I call the “Logical Explanation of Procrastination” to justify our unwillingness to learn. You may be familiar with some of the reasons this algorithm provides:
I have limited time since I have ______ (The blank is usually filled with too much work, family commitments, etc.).
I don’t know where to start.
Learning is an isolated pursuit and I don’t have the motivation.
I easily get distracted and don’t have the discipline
Insert your reason here
Additionally, most people think the solution is to either adopt the “hustle” mindset and burn the midnight oil (and the candle at both ends by extension) or quit their job and learn full-time.
These proposals aren’t sustainable. I learned machine learning (and deep learning eventually) while working full-time. I made a ton of mistakes and hit many walls along the way. Eventually, I came up with a sustainable approach to learning and I figured that might be useful to you. I call it
In 1952, Toyota invented a new system of manufacturing called Just-In-Time manufacturing. The core principle of this method is to bring the parts and components needed to build something exactly when it is being assembled and just when the parts are needed. This revolutionized manufacturing. Toyota saved a ton in warehouse space, eliminated wastage, and reduced costs.
Just-In-Time learning works the same way.
Learn something new exactly when you need it and bring together the resources you need as you are learning. This might seem a bit confusing. You could argue that you don’t do machine learning at work, that the topic you want to learn isn’t related to the research you do, or that you’re looking to learn machine learning to switch careers. What then?
This technique still works. How?
Just-In-Time learning is project-based. The project doesn’t need to be what you are doing at work (If it is, then that’s awesome). It just needs to be something you want to do.
Simply define a project you want to pursue, say an image stylizer or a speech synthesis system, and work back from the goal. Find papers that solve this problem and read them. Augment your understanding through repositories in GitHub that solve the problem or a sub-problem. Work back from the goal until you have every single step mapped out.
The key is to only collect and use resources that are directly relevant to your end goal. Ignore everything else.
This ensures that you don’t waste time searching and getting buried under a never-ending pile of resources. So, we have a starting point. How do we deal with the lack of time?
For this, I take a leaf out of James Clear’s book Atomic habits. There are 4 laws of habit formation:
Make it obvious
Make it attractive
Make it easy
Make it satisfying
Applying each to Just-In-Time learning, we get:
Make it obvious:
You only collect resources that help with the end goal, so there’s nothing unclear as to what you should ignore. Every step you need in this learning journey is obvious.
Make it attractive:
If the goal of your project is work-related, then the rewards are explicitly laid out. If on the other hand, you are looking to transition into a career in machine learning, the #1 thing that companies look for is proof of work. There’s NO bigger proof of work than a body of GitHub projects, open-source contributions, etc. Courses and transcripts only go so far. At the end of the day, companies want to know you can actually do the work hands-on.
Make it easy:
Work on it, piece by piece, regularly. Even if you only have 30 minutes a day to devote to this, that’s fine. Show up. Your learning compounds over time. By committing to a manageable chunk of time every day, you make it easy for yourself and remove the “lack of time” excuse from your quiver of reasons to not learn.
Do the math: 30 days x 30 minutes per day = 900 minutes of focused learning
Make it satisfying:
There’s an intrinsic satisfaction you get by just showing up. When you master a hard skill, you gain confidence, and each new thing you learn helps you learn harder things more easily. Learning is its own reward.
I use this approach to learn things even today and there’s rarely been a time when I felt frustrated or overwhelmed. I hope you’ll give this method a try and let me know your experience.
I’d love to know what challenges you face in your learning journey and what approaches have worked for you. Drop me a note.
🤖💪 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 →