A layer of a neural network has a few units. Each one has an activation: the number it passes on to the next layer. To stop the network leaning on any one unit, training makes units randomly not show up: each shows up with chance keep_chance, and this time the coin flips came out as kept (1 = showed up, 0 = didn’t). Write drop(activations, kept, keep_chance): what the layer passes on this time.
| unit 1 | unit 2 | unit 3 | unit 4 |
|---|---|---|---|
| 3 | −1 | 2 | 5 |
| unit 1 | unit 2 | unit 3 | unit 4 |
|---|---|---|---|
| 1 | 0 | 1 | 0 |