A recurrent neural network that serves as content-addressable memory. Given a partial or noisy input, it converges to the nearest stored pattern.
Neurons are fully connected with symmetric weights learned via Hebbian rule. The network minimizes an energy function, settling into stable attractor states.
What you're seeing
- Circular nodes = neurons (+1 or -1 state)
- Curved lines = synaptic connections
- Purple particles = activation flow
- Center glow = energy basin depth
Applications
- Pattern recognition & completion
- Associative memory retrieval
- Optimization problems