Thread by @svpino

Every recommendation to start with machine learning focuses on a few building blocks:

• Linear Algebra
• Calculus
• Statistics and probabilities
• Fundamentals of machine learning

But there's also a different way.

Let's talk about it: ↓

Many companies are dying to start applying machine learning to their businesses.

Believe me: I talk to many of them every week.

Their main problem: they don't know where to start or how to get it done.

If you are reading this, you are probably part of one of these companies.

Heck, for all I know, most employees out there work for a company that's in this situation!

The demand for machine learning professionals is enormous!

Unfortunately, those who want to start face a laundry list of things you "must-know" before you can use machine learning.

If you are a software developer with years of experience, you are probably not looking to start over or go back to school.

But you don't have to!

I have fantastic teammates working in the machine learning industry.

Many of them didn't go to college. Most didn't get a Ph.D. or Masters in machine learning or related fields.

But all of them are absolutely amazing at what they do!

We aren't researchers. We don't write papers. We don't create new techniques to beat state-of-the-art methods.

We use what's out there and apply it to bring value to the companies that need it.

We stand on the shoulder of giants and squeeze every ounce of it.

Many of my teammates started as I did:

We had a robust Software Engineering background.

We didn't know much about advanced calculus, but we learned how to build and use existing software.

And here is the best part:

The tooling around machine learning is evolving quickly!

A few years ago, we needed to write a lot of code to stitch many pieces together.

Today, much of the work is done by platforms that get data on one side and spit out a deployed endpoint to make predictions.

It sounds simple to take data, create a machine learning model, and deploy it.

I promise you it isn't.

Heck, it's very hard to do it correctly!

But today, for many problems, the difficult part has nothing to do with the skills you learn with your Ph.D. thesis.

Here are some of the challenges you need to worry about when building these systems:

• Scalability
• Monitoring
• Versioning
• CI/CD/CT

Do you know who is good at these Software Engineers!

Do you know who else? Platforms that solve this for us!

Let's talk about a platform that you can start using right now, and your work will never be the same: @abacusai.

Here is how it works:

First, you start with the problem you want to solve.

The platform supports many different use cases out of the box.

Here is a list of some of these use cases:

• Personalized recommendations
• Financial metrics forecasting
• Image classification
• Predictive analytics
• Customer churn predictions

The list goes on and on!

Most of the things you'll have to learn will be around identifying the correct use case and putting together the data to solve it.

The good news?

Many of these problems are the same from company to company!

Most companies I talk to want to solve the same issues!

Second, after you know the problem you are solving and have the data for it, you'll upload it to @abacusai's platform.

Getting the correct data and transforming it will take some time.

But there are guidelines, and they will help.

Third, now that data is in the platform, you can train a model.

It's a button.

Just a button that you press to train the model!

No, you don't need to understand the specifics of the model, how it works, or the math behind it.

You press the button and your model trains.

Let me take a second here to make something clear:

To use @abacusai effectively (or any other machine learning platform for that matter,) you need skills.

It's a tool for professionals, not average users.

But the tool hides a lot of the complexity!

Not having to deal with that complexity frees us to focus on the things that matter:

• Are we solving the right problem?
• Is this the best way to solve it?
• Do we have the correct data?
• Is the data clean and representative?
• Are there any biases?

Let's get back on track:

After we train the model, we can deploy it.

I know, it sounds simple: "deploy it."

If you have done this before, I'm sure you know what I'm thinking: Deploying a model is hard work!

But @abacusai does it for us!

To summarize the process :

1. Choose the use ca e.
2. Upload the data.
3. Tr in a model.
4. Deploy it.

The tool worries about a lot of the complexity, and we focus on what matters to us!

But wait a second!

I know what many of you will say (because you have told me before):

"Clicking buttons doesn't make you good at machine learning!"

I agree, but who cares?

Most people care about delivering value.

They worry about solving problems that improve the lives of their customers.

There are many ways to accomplish this, and using tools like @abacusai is a smart strategy to get there.

This is a different path for those who want to use their skills to solve problems.

If you are a software engineer, I'd recommend considering this approach.

It's not better. It's not worse.

It's just a different and effective way to accomplish your goals.

As the field matures, we will continue moving to higher levels of abstractions.

I wrote a training loop to train a neural network from scratch. Never had to do that again.

Practicing machine learning five years from now will look very different from what we are doing today.

These are exactly my arguments.

As we make progress in the field, "building blocks" move up the ladder: what you need to understand is at a higher level than what it was before.

https://twitter.com/abdullah_mufti/status/1495788491220955146?s=20&t=dw-plU5U8LodkCv4dTomlw

I'm not a mathematician.

I also don't hate math, but it's definitely not what I read in my downtime.

I have not seen any evidence that machine learning is limited to mathematicians or people who are good at math.

We are far, far away from getting to the point where everything is done by "pushing a button."

However, this doesn't mean that we need to start from the same place we did 5 years ago.

We have made a ton of progress, and we continue to do so.

I really like this analogy.

https://twitter.com/TrentSeigfried/status/1495857159824986120?s=20&t=dw-plU5U8LodkCv4dTomlw