0.3.0

Minor version removing "model" module to allow more complex architectures using maps and the library nnet.
A synchronisation step added at cortex/neuron start to shutdown neurons which create infinite rcc loops or do not influence any of the network outputs.

Performance improved using transactions

Dependency with edb library removed.
Now the application interfaces directly with mnesia improving performance.

Recurrent networks stable (without layer index)

Fix issue with deadlocks in recurrent neurons.

Network without layers

New modules: network, nn_node, link which control the network simplifying the structure removing the layer concept (similar to digraph: https://erlang.org/doc/man/digraph.html).

Transformation module not needed as the network concept is created. A network can be modified multiple times without the need of saving each time in mnesia.

Link module is extracted from neurons, so to modify a link, it is not needed to write two parallel transactions to the neuron input and output. Only one to the network.

Elements module merged into cortex, neuron and network.

Stable back propagation

Stable back propagation with good results

• Activation functions
• Aggregation functions
• Initializes