Machine Learning: The Bare Math Behind Libraries
summary
Starts from linear regression with a concrete objective function and implements gradient descent to minimize squared error, showing code run in Octave and reporting convergence after 1,000 iterations to fit price-versus-issues data. Extends to a feedforward neural network with 200 hidden neurons trained to reproduce a Batman-like logo (works on the training grid but generalizes poorly to a denser 20k×5k grid). Covers unsupervised rules: Hebbian updates, winner-takes-all, winner-takes-most and Kohonen self-organizing maps with practical stability and clustering tradeoffs.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:29Definitions
- 13:05In Practice
- 14:55Gradient Descent
- 19:00Neural Network
- 23:47Unsupervised Learning
- 35:55Winner Takes All
- 39:55Winner Takes Most
- 40:53SelfOrganizing Map
- 41:21Solution Space
- 41:57Conclusion
- 42:50Outro
related talks
53mMachine Learning: The Bare Math Behind Libraries
51mMachine Learning Exposed Introduction to Machine Learning
33mMachine Learning explained to humans
49mMachine Learning Exposed: Introduction to Machine Learning
44mMachine Learning Exposed
51mMachine Learning Explained to Humans
51mMachine Learning Exposed Deep and Reinforcement Learning
45mFrom Ancient Greeks to Modern Geeks - Basic Machine Learning Algorithms