Hands-On One-shot Learning with Python

Visualizing parameters

Neural networks learn through gradient descent, but what do they learn? The answer is parameters, but we are looking to understand what those parameters mean. In the following diagram, if we look at the first few layers, we will see simple and comprehensible extracted features, such as edges and interest points, whereas deeper layer features are more complex. For example, if we look at the last layer in the following diagram, we will observe that features are indecipherable compared to the initial layer features. This is because as we go into deeper layers, more information-rich features are being extracted through various matrix operations. This enables high-dimensional information to be compressed into the low-dimensional ...