Abstract
The storage and generalization abilities of a multilayer neural network with a number of hidden units scaling as the input dimension were characterized. The mapping from the input to the hidden layer was realized by symmetric Boolean functons with L inputs. The storage capacity was found to be proportional to the logarithm of the number of these Boolean functions divided by L.
| Original language | English |
|---|---|
| Article number | 078101 |
| Pages (from-to) | 078101/1-078101/4 |
| Journal | Physical Review Letters |
| Volume | 87 |
| Issue number | 7 |
| State | Published - 13 Aug 2001 |