A network is a stack of blocks. In each one a layer computes a change = layer_weight × the signal, number by number. Write block(signal, layer_weight): what this block passes on to the next. At the start of training every layer weight is near 0, and a deep network stacks 50 of these blocks.
| 2 | −0.5 | 1 |
| 0.2 | −0.05 | 0.1 |