Short interactive experiments
Energy, memory, probability, and hidden variables.
Each experiment isolates one central idea behind Hopfield networks, Boltzmann machines, and restricted Boltzmann machines. Use one during a lecture, or let people explore after scanning a QR code.
+1J-1
3. Boltzmann probability
Energy becomes a distribution over states.
2. Hopfield associative memory
A learned energy minimum can repair a corrupted pattern.
1. Hopfield to Boltzmann
Replace a threshold by a probability.
4. Restricted Boltzmann machine
Hidden causes create correlations
Tune a visible-hidden model for Anna, Lukas, and Bruno. Inspect probabilities and sample one Gibbs step.
0p1
Lecture companion
Work along the slides
Three short interactions follow the lecture: stochastic updates, Boltzmann probabilities, and positive/negative-phase learning.