Building & Optimization of Neural Networks from Scratch with Numpy
summary
Builds a working neural network from scratch using a NumPy-only implementation, coding sigmoid activation, its derivative, and the backpropagation equations, then trains with stochastic gradient descent (learning rate ~0.1) for 10,000 epochs. The example is a movie‑recommendation toy dataset with two inputs (Mary/John ratings), one hidden layer (hidden size 4) and one output; weights initialize randomly, training converges and the final sigmoid output crosses 0.5 to recommend the movie.
AI-written summary · about
chapters (11)
- 0:00Neural network tutorial intro
- 1:07Movie recommendation case study
- 13:25Linear decision limits
- 19:59Multi-layer networks
- 27:14Forward propagation in NumPy
- 35:56Parameter initialization
- 45:07Forward function definition
- 59:20Cost function calculation
- 1:02:04Backpropagation explained
- 1:16:03Training over epochs
- 1:21:14Summary and next steps
related talks
25mIntroduction to Neural Networks
04mNeural Networks Demystified: Part 1
43mA Gentle Introduction to Neural Networks
01h 50mNeural Networks: Walkthrough
05mNeural Networks Demystified: Part 6
43m(Deep) Learn You a Neural Net for Great Good!
01h 04mWriting a Neural Net from Scratch
07mNeural Networks Demystified: Part 3