Release v9.0.0

Issue #28 Enhancement of the readout build progress notification events architecture.
Issue #29 More code examples and better structured menu of demo application.

Release v8.0.0

The redesign of the Readout Layer, introduction of the Cluster Chain as the standard computation unit for the Readout Unit and One Takes All group. Enhanced data bundle folderization. Enhanced and revised code comments.

Release v7.3.1

This release contains the critical patch of v7.3.0 (release v7.3.0 was discarded to avoid ).
Introduction of Double Twisted Toroid interconnection schema. Small enhancement of network cluster members weighting. Refactoring of Activations. Added the SoftMax activation. Inherited doc-comments where possible. Unmodifiability option of Interval and sharing of commonly used static intervals.

Release v7.2.0

Issue #25 StateMachineDesigner 2nd level cluster computation, Crossvalidation element, Regression noncritical bugfix

Release v7.1.0

Issue #24 - Enhancement of the networks cluster computation (ReadoutUnit)

Release v7.0.0

Main changes

• Redesign of spikes coding (A2S coders). Horizontal, Vertical or Forbidde spiking input processing (InputEncoder)
• Issue #23 PredictorsProvider redesign and enhancement
• Bidir modes
• More examples and small enhancements of the StateMachineDesigner

Release v6.0.0

RCNet library on .NET Standard 2.0
Demo console application on .NET Core 3.1

Release v5.2.1

Added empty connection schema to enable prime spiking neuron pools. Use is demonstrated in the BeetleFly demo case.

Release v5.2.0

This release brings several enhancements, simplifications and it also fixes several bugs:

• an improvement of analog hidden neuron's capability to fire a spike. Up to now, firing event at analog hidden neuron was defined as a situation, when the difference between current state(T) and previous state(T-1) of analog activation exceeds specified threshold and the previous activation state was fixly considered as the state(T-1). But now, it is possible to define the "deepness". So if we define deepness=2 then half of neurons still uses state(T-1) as the previous activation state, but another half of neurons uses state(T-2) as the previous activation state. For deepness=3, one third of neurons uses state(T-1), another third of neurons uses state(T-2) and another third of neurons uses state(T-3). ... etc.
  This new feature can be very powerful as it is now demonstrated in CricketX demo case
• new optional "binary" component of the input spike code and improved distribution of spiking input connections.
• synapse performance improvement. Now it is enabled "aggressive inlining" when data from synapse is fetched and synapse has also direct access to hidden neuron's data (data access through neuron's interface INeuron was eliminated)
• SignalingRestriction removal. Never really used feature was removed for the simplification
• bugfixing of CsvDataHolder (correct work with input stream)

Release v5.1.0

This release brings a possibility to define steady data fields within the input pattern context.
It also contains bugfixing and several small enhancements related to neural statistics and probabilities mixing.