A Fireside Chat with Turing Award Winner Geoffrey Hinton, Pioneer of Deep Learning
summary
Describes why connectionist learning wins: backprop struggling until unsupervised pretraining (circa 2005) plus GPU use (from 2007) enables deep nets; speech models reach production (Android, 2012) and vision explodes with AlexNet (2012). Explains Boltzmann machines and wake‑sleep dreaming/unlearning math, argues top‑down reconstruction reduces adversarial failures, proposes capsules and coordinate‑frame routing to generalize better, and predicts neural nets eventually replicate cognition (claims 99.9% Bayesian confidence).
AI-written summary · about
chapters (5)
- 0:00Introduction
- 0:55What is a Neural Network
- 6:10When does a Neural Network start to work
- 15:10Neural Networks
- 21:18Learning the Brain
