Basic McCulloch & Pitts Neuron
| A | B | A v B |
|---|---|---|
| 1 | 1 | 1 |
| 1 | 0 | 1 |
| 0 | 1 | 1 |
| 0 | 0 | 0 |
| C | D | C v D |
|---|---|---|
| 1 | 1 | 1 |
| 1 | 0 | 1 |
| 0 | 1 | 1 |
| 0 | 0 | 0 |
| A v B | C v D | (A v B) ^ (C v D) |
|---|---|---|
| 1 | 1 | 1 |
| 1 | 0 | 0 |
| 0 | 1 | 0 |
| 0 | 0 | 0 |
| (A v B) ^ (C v D) | ¬(A v B) ^ (C v D) |
|---|---|
| 1 | 0 |
| 0 | 1 |
Autoencoder
Auto Encoders learn to “reproduce” their own input after passing it to a lower dimensional layer.
Recurrent Neuron Network
- In RNN, output is fed back into hidden layer, serves as “memory”
- Useful to analyze sequences of data, e.g. speech recognition
Convolutional Neuron Network
The output size is:
$$ \frac{N-F}{stride} + 1 $$
- F: Size of filter
- N: Size of original input image
- stride: step size
Thus:
$$ \frac{15-3}{1} + 1 = 13 $$
Because there are 3 filters, the output size is 13 x 13 x 3$$ \frac{15-3}{2} + 1 = 7 $$
Because there are 3 filters, the output size is 7 x 7 x 3To preserve the size of the output for the stride 1 filter, we need to add the zero-paddings with:
$$ p = \frac{F - 1}{2} $$$$ p = \frac{5 - 1}{2} = 2 $$
Thus we can perserve the size of output by padding with 2 pixel border to the input image