is Chained-Cubic-Cell (previous abbrev.) or Chained-Cauchy-Cell (current abbrev.).
An implementation of probabilistic multi-layered perceptron.
All connection weights and biases are random variable depending on Stable distribution in this model.
The signal propagation gets randomly determined and the unit randomly fires.
Cauchy distribution is employed for current implementation.
Their units can take only binary states, 0 and 1, meaning "not fired" and "fired,"
similat to general multi-layered perceptron.
The common of learning law is similar to the BP and RTRL for general (recurrent) neural networks.
They however employ several improved techniques such as sign propagation for binary unit.
This framework depends on only Accelerate, Metal and Core Data.
The computation is accelerated by GPGPU and their learnt relationships are managed by ORM.