Neural Networks Demystified: Part 3
summary
Builds on a simple Python network and shows exactly how cost is measured as the sum of squared errors with a 1/2 factor then minimized by changing weights rather than brute force search. Demonstrates timings: ~0.4s for 1,000 weight tests, ~40s for two weights, ~11 hours for three, and an astronomic multi‑quadrillion‑year estimate for nine. Introduces gradient descent and stochastic gradient descent, claims gradient descent cuts a 10^27 evaluation problem to under 100 iterations.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:41Cost Function
- 1:07Minimizing a Cost Function
- 2:29The Full Complexity of the Problem
- 4:01Numerical Estimation
- 5:13Gradient Descent
- 6:03The Convex Nature of Quadratic Equations
related talks
25mIntroduction to Neural Networks
52mNeural Networks Walkthrough
01h 50mNeural Networks: Walkthrough
58mNeural Networks: A Primer
01h 25mBuilding & Optimization of Neural Networks from Scratch with Numpy
01h 02mNeural Networks: A Primer
28mDemystifying Deep Neural Nets
51mMachine Learning Explained to Humans