- 2071 Magh · 3 marks
Hopfield vs Kohonen network
Short answer
Hopfield network is a recurrent auto-associative memory with symmetric weights used for pattern completion. Kohonen SOM is unsupervised competitive learning for clustering and topology-preserving mapping. Hopfield iterates to stable states; SOM finds best matching unit.
Compare Hopfield versus Kohonen network.
| Point | Hopfield network | Kohonen network (SOM) |
|---|---|---|
| Type of learning | Hebbian-type, effectively supervised storage of patterns | Unsupervised, competitive learning |
| Architecture | Single layer, fully connected, recurrent | Input layer and a 1-D/2-D output (map) layer, feed-forward |
| Connections | Symmetric weights, no self-loops | Each input linked to every output node; lateral competition |
| Purpose | Auto-associative memory, pattern completion, optimisation | Clustering, topology-preserving mapping, dimensionality reduction |
| Weight setting | Computed once: W = Σ pᵀp (diagonal 0) | Updated iteratively: w ← w + α(x − w) for winner and neighbours |
| Neuron output | Bipolar/binary, threshold function | Winner-takes-all (closest weight vector wins) |
| Recall | Iterates until a stable state (energy minimum) | Finds best matching unit for an input |
| Capacity/limit | About 0.15 N patterns; spurious states | Needs choice of map size, α and neighbourhood |
✓ Answer checked by Claude Opus 5.5 (high effort)
Related questions
- Hopfield network and its steps · Artificial Intelligence, chapter 6
- Neural network; supervised vs unsupervised · Artificial Intelligence, chapter 6
- Context switching; distributed vs network OS · Distributed System, chapter 3
- Network OS vs distributed OS · Distributed System, chapter 3
- Crossbar switch vs multistage network · Computer Organization and Architecture, chapter 8
- Compare MOCB/BOCB and ABCB/puffer SF6 · Switchgear and Protection, chapter 6