Remember when you were learning math in high school and thought to yourself:

“When am I ever going to use this?”

Well you’re reading a newsletter about machine learning, which means that time is:

I know. Not everyone considers the math the “life of the party” (unless you’re like me and love it).

In the world of machine learning, math is like that quiet friend who suddenly reveals they can hack any system or speak 10 languages.

It’s unexpectedly badass.

And in the case of ML and math, well let's just say they go together like Romeo and Juliet.

Alright, let’s get real for a second. You might be thinking, “Can’t I just use pre-built Python libraries and let the computer do all the math?”

Sure, you could. And eventually, that’s how you should do it.

But that’s like thinking you’re a Formula 1 racer because you’re good at Mario Kart.

Anyone can turn a key and make a car go. But it takes a real engineer to understand what’s going on under the hood.

And that’s exactly how you should be thinking…like an engineer.

In ML, math is your tool set — it lets you tune the engine, optimize performance, and even design new models.

Without math, machine learning would be about as useful as a screen door on a submarine.

Thus its no wonder ML feels like this towards math:

Now you don’t need to be the reincarnation of Einstein in order to grasp the math for ML.

Instead, think of the tortoise in a hare race…taking steps one at a time.

Eventually, the math pieces will fall into place, and everything will start to click.

The holy trinity of math for machine learning is a way to describe the three core fields you need to know: Linear Algebra, Calculus, and Probability & Statistics.

Think of them as Marvel’s Avengers of ML.

Linear Algebra builds the world.

Calculus helps fight bad choices.

Probability makes predictions — sometimes wrong, but always confidently.

I’ll take you through each core field more comprehensively in the future, but for now, I just wanted to introduce you to this awesome team that makes ML what it is.

There is a type of math that’s lesser known (but not lesser in stature!) than the 3 mentioned above: Information Theory. Although in this particular post, I will focus on the 3 above, this is another math I will cover in the future (just giving you a heads up!).

If your data is the story, linear algebra is the language it's written in.

It’s all about working with vectors, matrices, and tensors—fancy names for rows and columns of numbers (imagine a Google Spreadsheet of numbers!).

Almost every dataset in machine learning, whether it's an image, a spreadsheet, or a text document, is represented as a matrix.

Basically, its the IKEA instruction manual for how data gets stored, rotated, and squished into shapes your algorithm can actually read.

Linear Algebra helps us manipulate our data.

It's how we scale features, rotate images, and flatten everything into a format a machine can understand. Without it, your data is just a mess of numbers

But with it, it's an organized, ready-to-use dataset that will train your model to infinity and beyond.

And here's what's happening under the hood: dot products multiply and sum rows and columns together, which is the core operation behind every neural network layer.

Eigenvalues and eigenvectors power PCA (Principal Component Analysis), compressing your data without losing what matters.

And matrix multiplication is how data flows through every single layer of a neural network.

Every time a neural network makes a prediction, it's doing thousands of matrix multiplications under the hood.

Calculus is the driving force behind a model’s ability to learn.

It's all about finding the rate of change and the slopes of things. In machine learning, this translates to finding the best way to make your model smarter.

Its kinda like your GPS constantly calculating the best, shortest route to your destination.

The “Gradient Descent” is more than just the name of my newsletter. It’s named after a concept from calculus!

It tells the model which direction to move to get closer to a correct answer, so that it actually makes educated advances, rather than random guesses.

Here's the deeper picture: derivatives measure how much the model's error changes when you tweak a single parameter.

The chain rule is the backbone of backpropagation, the process by which neural networks learn layer by layer, tracing errors all the way back to the source.

Loss functions are the "score" that calculus is constantly trying to minimize.

Backpropagation is just the chain rule applied thousands of times — calculus doing the heavy lifting so your model can self-correct.

Without calculus, your model would be like a drunk dude trying to play darts.

In life, we often feel uncertain. Confused. Even lost.

In ML, that feeling of uncertainty won’t go away. Fortunately, there’s a way for us to handle uncertainty: Probability & Statistics.

This is where we figure out how likely something is to happen, based on the data we have.

Probability & Statistics helps us evaluate a model's performance (Is it accurate? Is it guessing correctly?), and it’s the foundation for many algorithms, like Naive Bayes.

Going deeper: distributions (like normal or Bernoulli) describe what your data looks like and what outcomes are likely. Bayes' Theorem is how models update their beliefs when new data arrives.

Understanding overfitting vs. underfitting is a statistical reality check on whether your model actually generalizes or memorizes the answers.

A model without statistics is just pattern-matching with confidence it hasn't earned.

Remember our drunk dude from earlier? ML without probability is like him playing poker, betting his car on a 2 and a 7 because it “feels lucky”.

Ouch.

Time to get into the math with actual practice. Click below!