Nathan Lamberson’s Reviews > Why Machines Learn: The Elegant Math Behind Modern AI > Status Update

Foundational backstory of the start of forming the nearest neighbor algorithm

Finished probability chapter, yay. I hate prob&stat. Summary on p. 140 to help review if needed.

Probability to prove who wrote disputed federalist papers based off word distribution. Kinda cool

Probability & Bayes theorem, my biggest nightmare from college, yay.

How Least Mean Square (LMS) algorithm was made and its use cases (p. 91) and a good diagram of how it works (p. 87)

Adaptive filters and how dial up works by screening out the error signal

Lots of proofs on how Perceptons function. See p. 57 for start of proofs on their updating always working and how it wont run infinitely.

Lots of math and proofs on finding a hyperplane and shifting the weights. Algorithm on p. 51. Talks about setting upper and lower bounds for determining how many times we should re-calculate, since for complex problems you could run forever and ever searching for an answer

Dot products and matrix multiplication, which honestly lost me.

P. 46 has an important equation and diagrams for understanding all that connection to the Perceptron, which makes some sense at least. The vector for weight, w, will always have a perpendicular line to itself (itself being hyperplane) that is what classifies this outputs as -1 or 1.

Vectors (magnitude and direction) and scalars (quantity, just a number)

Def learned this before in Physics or calc in college. It clicked again on pp. 29-30 with the graphs. The Cartesian coordinates we def used before on p. 33. Vector math is super simple too, see p. 32.