What math does ML need?
ML rests on three areas of math: linear algebra (vectors and matrices, which are how data and models are represented), calculus (gradients, which are how models learn), and probability and statistics (which are how we reason about uncertainty). You need working intuition, not a degree.
Why it matters
Without the math, ML is a black box you can only copy from tutorials. With it, you understand why a model trains, why it fails, and what the metrics mean. You do not need to derive everything by hand, but the intuition is what separates practitioners from people pasting code.
What to learn
- Vectors, matrices, and matrix multiplication
- Dot products and what they represent
- Derivatives and gradients as slopes
- The chain rule, which powers backpropagation
- Probability basics and distributions
- Mean, variance, and standard deviation
- Why all of this maps onto data and models
Common pitfall
Trying to master all the math before touching any ML, and stalling for months in abstract theory. Learn the intuition for each concept as the ML topic that needs it comes up — gradients when you hit training, linear algebra when you hit tensors. Just-in-time math keeps you moving and makes it concrete.
Resources
Primary (free):
- 3Blue1Brown — Essence of linear algebra · video
- 3Blue1Brown — Essence of calculus · video
- Khan Academy — Statistics & probability · docs
Practice
By hand or in NumPy, multiply two small matrices and compute a dot product, then explain what the result represents. Compute the derivative of a simple function and describe it as a slope. Done when you can connect each operation to its role in training a model.
Outcomes
- Perform and interpret matrix multiplication and dot products.
- Explain gradients and the chain rule intuitively.
- Reason about distributions, mean, and variance.
- Connect each math concept to its use in ML.