Thread by @svpino

"You can't use an algorithm unless you understand how it works."

That's what many people say. But I don't believe it.

This is how you can build expertise: ↓

We all learn new things in different ways.

Personally, I'm a huge proponent of learning on-demand:

• Start with a problem
• Try to solve it
• Incorporate new knowledge as you go

Almost every time:

I start using new techniques with a very superficial understanding of how they work.

Sometimes, I only know they *do* work but have no idea how.

As I progress and find problems, I'm forced to understand more about what I'm doing.

At a high level, this is what I do:

Find a solution that works → Learn how it works

Here is a specific example:

I trained my first model to classify Dogs and Cats before I understood how Convolutional Neural Networks worked.

• I solved the exercise
• Found a problem
• Started digging for a solution
• Learned how to do it
• Fixed the problem
• Repeat

This is important:

• Using something doesn't make you an expert on it.

Copying code to solve a specific problem doesn't automatically qualify you to push that to production.

You have to earn that right.

This is where people get stuck:

Many assume that those who start with a hands-on approach are skipping ahead. They are dangerous, and their solutions are mediocre.

This is a poor generalization.

Knowing the theory doesn't qualify you to drive a car. Knowing the physical aspects of driving doesn't qualify you either.

You need a combination of both.

The order that you follow to learn doesn't matter.

We all have to gain the privilege of sharing our work with the world.

You don't get extra points for learning theory first. You don't get extra points for learning theory as you go.

• We all need to show our value.
• We all need to build trust with our teams.

What's the takeaway?

• Find the path that works for you and focus on building expertise.

Over the years, the approach I explained here has proven to be exactly what I need.

Find yours.

Every week, I post 2 or 3 threads like this, breaking down machine learning concepts and giving you ideas on applying them in real-life situations.

You can find more of these at @svpino.

If you find this helpful, stay tuned: a lot more is coming.

That's an excellent opportunity to dive deeper and get a better understanding of what you are doing.

Like peeling an onion, you start layer by layer until you find the problem.

https://twitter.com/shakilmith/status/1442816779525394434?s=20

People assume that "writing code" is the only necessary requirement to build software.

It is not.

You need to gain your team's trust before they let u push code to production.

You gain trust by showing good work, not by "understanding everything."

https://twitter.com/ankit_JdS/status/1442811998756958217?s=20

There's a subtle—but fundamental—difference in the way we are using the idea of "understanding an algorithm" here.

My argument is that you can use algorithms that you understand their expected result without having to understand how they arrive at it.

https://twitter.com/zakhele_madi/status/1442904349827481600?s=20