At the begginer stage, the goal is to master multivariate differentiation, integration, and vector fields, while building a strong geometric intuition for how gradients move through space.

"Calculus" by Gilbert Strang

Why it's great: Like his famous linear algebra textbook, Strang focuses on the why and the big picture rather than tedious algebraic trickery. It is open-source, highly accessible, and provides the exact geometric intuition needed to understand optimization landscapes.

"Thomas' Calculus" (or "Stewart Calculus")

Why it's great: These are the gold-standard engineering calculus textbooks. You don't need to read them cover-to-cover, but the chapters on Multivariable Calculus, Partial Derivatives, Gradients, and the Jacobian/Hessian matrices are mandatory practice for understanding how deep networks compute errors.

And to cover all

"Mathematics for Machine Learning"By Marc Peter Deisenroth, A. Aldo Faisal, and Cheng Soon Ong

What it covers: It acts as the ultimate bridge. It strips away the physics-heavy parts of standard calculus and delivers exactly what an AI researcher needs: Vector/Matrix gradients, Jacobians, Hessians, Taylor series approximations, and the math behind Backpropagation and Gradient Descent.Why it's required: Standard calculus books teach you how to take the derivative of a scalar function. This book teaches you Matrix Calculus—how to take the derivative of a matrix loss function with respect to a vector of billions of weights.