Derived types

The derived types provided by the ATHENA library can be separated into subcategories. The prominent derived type is

network_type

This is the container for a neural network and all its layers.

Layer Types

The followin layer types are available:

• Convolution layers
• Fully connected layers (aka dense layers)
• Input layers (automated)
• Normalisation layers
• Pooling layers
• Regularisation layers
• Reshaping layers (automatic)

Optimiser and associated types

There exist a set of derived types that are used to store the optimisation parameters and procedures, in addition to the training metrics. These types are:

• base_optimiser_type
• sgd_optimiser_type
• rmsprop_optimiser_type
• adagrad_optimiser_type
• adam_optimiser_type
• metric_dict_type