definition

Cascade correlation is a supervised learning architecture that can be used to build a minimal multi-layer network topology. Its advantage is that users do not need to worry about the topology of the network, and its learning speed is faster than traditional learning algorithms.

Related algorithms

The cascade correlation algorithm is implemented in the following way:

1. Start with a minimal network, containing only input and output layers, with both layers fully connected;
2. Train using a traditional learning algorithm, closing the link at the output unit until the error stops increasing;
3. Generate so-called candidate units, where each candidate unit is connected to an input unit and a hidden unit;
4. Try to resolve the correlation between the remaining errors through training;
5. The candidate unit with the largest correlation is selected, its input weights are frozen and added to the network.

Repeat the above steps until the total network error is lower than the estimated value.