There are two very different jobs hiding under the word “AI.” AI research invents new model architectures and needs deep mathematics. AI engineering builds useful systems with models that already exist — and needs far less. Most jobs, and almost every beginner project, are engineering. So before you panic about math, decide which path you are actually on.
The short answer
To use machine learning — training models with libraries, building AI features, doing data science — you need a working grasp of four areas, none of them exotic: linear algebra, a little calculus, probability, and statistics. You do not need to derive proofs; you need enough intuition to understand what the tools are doing and why a result looks wrong. To research new methods, you need those same areas at much greater depth. This guide is about the first path.
The four areas that actually matter
| Area | What to understand | Why it matters for ML |
|---|---|---|
| Linear algebra | Vectors, matrices, dot products, matrix multiplication | Data, weights, and embeddings are all vectors and matrices; it is the language of ML |
| Calculus | Derivatives, gradients, the chain rule (conceptually) | Training works by gradient descent — following slopes downhill to reduce error |
| Probability | Distributions, expectation, conditional probability, Bayes | Models predict likelihoods and reason about uncertainty |
| Statistics | Mean, variance, sampling, correlation, evaluation metrics | Measuring whether a model is actually good, and avoiding being fooled by data |
Linear algebra is the one to prioritise
If you study one area first, make it linear algebra. Every dataset is a matrix, every model's parameters are vectors and matrices, and every embedding — the numbers behind modern AI and large language models — is a vector in space. You do not need to invert matrices by hand, but you should be comfortable with what a vector is, what a dot product measures, and what matrix multiplication does. That intuition makes neural networks stop looking like magic.
You can learn the math just-in-time. Many successful ML engineers build working projects first with libraries, then deepen the specific math behind each technique as they hit it. Understanding beats memorising formulas you never use.
What you can safely skip (at first)
- Hand-deriving backpropagation — understand the idea (gradients flow backwards); let the framework do the calculus.
- Heavy proof-based real analysis and measure theory — research territory, not engineering.
- Memorising distributions and formulas — know what they mean and look them up when needed.
- Optimisation theory beyond gradient descent — useful later, not a starting requirement.
A sensible order to learn it
- Linear algebra intuition — vectors, matrices, dot products, matrix multiplication.
- Probability and statistics basics — distributions, expectation, variance, evaluation metrics.
- Calculus for ML — derivatives, gradients, and the chain rule, conceptually.
- Apply immediately — implement each idea on real data so the math attaches to something concrete.
Build the foundation the practical way
The mistake is treating ML math as a wall to climb before you are allowed to start. It is not — it is a toolkit you sharpen as you build. The fastest path is to learn the reasoning and the engineering together, each making the other click.
AI Engineering Essentials teaches AI from the ground up — from NumPy, data, and the math intuition you actually need, through neural networks, transformers, embeddings, RAG, and deployment — with runnable code and exact output on every page, and no PhD-level math assumed.
AI Engineering Essentials
A project-based path from data foundations to modern AI engineering, with code/output traces and practical checkpoints.
Buy the PDF for $25 Preview pagesIf you want to shore up the underlying reasoning first, Math & Logic Essentials rebuilds arithmetic, algebra, logic, probability, and statistics from first principles — the foundation that makes the ML-specific math feel obvious instead of intimidating.
Math & Logic Essentials
A friendly foundation for arithmetic, algebra, functions, geometry, statistics, sets, truth tables, conditionals, and proof habits.
Buy the PDF for $20 Preview pagesFrequently asked questions
Do you really need math for machine learning?
To use machine learning, you need a working intuition for linear algebra, basic calculus, probability, and statistics — not proof-level mastery. To invent new methods (research), you need that math far more deeply. Most people only need the practical, intuition-level version.
What math is most important for machine learning?
Linear algebra. Data, model weights, and embeddings are all vectors and matrices, so it is the language ML is written in. Probability and statistics come next, followed by enough calculus to understand how training works.
Can I learn machine learning if I'm bad at math?
Yes. Many ML engineers build real projects with libraries while learning the underlying math just-in-time, as each technique requires it. Strengthening your basic math and logic foundation first makes the journey much smoother, but it is not a gate that keeps you out.
How much calculus do I need for deep learning?
Conceptually: derivatives, gradients, and the chain rule — enough to understand that training follows slopes downhill to reduce error. The frameworks compute the actual derivatives for you, so you rarely do calculus by hand.
Is statistics or linear algebra more important for AI?
Both matter, but linear algebra is the more universal foundation for modern AI because everything is represented as vectors and matrices. Statistics is essential for evaluating models honestly and reasoning about uncertainty.