You do not need to be a pure mathematician, but you should:
In Linear Algebra and Learning from Data (henceforth LALD), Gilbert Strang—Professor of Mathematics at MIT—takes on the challenge of reorienting linear algebra toward data. The book’s central thesis is that . Every neural network weight update, every principal component, and every least-squares fit is ultimately a linear algebraic operation. Strang G. Linear Algebra and Learning from Data...
. It’s about understanding the and how we can manipulate them to find patterns, reduce noise, and make predictions. Core Pillars of the Text You do not need to be a pure
But the novelty is the exercises . You won't find abstract proofs about vector spaces. You'll find coding prompts (in Julia, MATLAB, or Python) asking you to factor matrices and visualize projections. You won't find abstract proofs about vector spaces
To appreciate the book, one must understand the three major philosophical pivots Strang makes.
Strang famously writes: "The fastest way to compute the SVD of a trillion-by-trillion matrix is to avoid computing it exactly."