Neural Networks - Convolutional
Convolutional neural networks (convnet or cnn) are deep, feed-forward artificial neural networks, most commonly applied to analyzing visual imagery.
Most CNNs have one main pair of operations:
- Convolution: slide a filter across an image e.g. if you are looking for vertical lines, you slide a vertical line filter across the image, you get back feature maps showing where the vertical lines are.
- Pooling: helps to everything to fit in memory and removes insensitivities e.g. vertical lines close to each other.
What the convnet architecture usually looks like:
Image > Convolution > Pooling > Convolution > Pooling > Fully-connected > Output
CNNs usually learn the training data in layers in the following way:
- 1st Layer: detects edges
- 2nd Layer: detects corners and curves
- 3st Layer: detects square, triangle and circle
- Rest of the Layers: detects special characteristics of the use-case