The Hopfield Network
- A Hopfield unit is a thresholding unit.
- Units are connected by weighted bidirectional connections.
- It can be discrete or continuous, but we'll use discrete.
- If a set of neurons is turned on in a cycle, and the system
is run, it will settle into a stable state or an oscillator.
- Example, 5 units with one pattern. Show auto-associativity.
- A net of N units can store N patterns all differing by more than
one unit.
- There is a Hebbian calculation to store the patterns, see
Hopfield Wiki.
- This is based on statistical mechanics and is a spin-glass model.
- How would this be used for machine learning?
- Quick store, use similarity, but better than city block.
