Lecture 1.4 — A simple example of learning [Neural Networks for Machine Learning]

3-Second Blueprint (English)

1. A two-layer neural network connects raw pixel intensities directly to class units, where active ink pixels vote for candidate shapes.
2. Hinton weight maps visualize thousands of connection strengths as positive or negative blobs evolving from noise into digit templates.
3. Pure template matching fails on diverse handwriting styles because robust recognition requires extracting intermediate local features.