Artificial Neural Networks for Anomaly Detection
Description
Training of general classification and regression neural networks using
gradient descent. Special features include a function for training replicator
neural networks and a function for training autoencoders. Multiple activation and
cost functions (including Huber and pseudo-Huber) are supported, as well as L1
and L2 regularization, momentum, early stopping and the possibility to specify
a learning rate schedule. The package contains a vectorized gradient descent
implementation which facilitates faster training through batch learning.