Soevnn is an acronym for Self-Organising Estimated-Variance Neural Net. It is consists a heterogeneous population of neurons connected in a directed cyclic graph. Instead of being pre-organized into specific layers, the neurons are organized into structures, types, and clusters. The neurons form connections probabilisticly based on a variety of factors. Neurons will also break less useful connections. Each neuron learns short-term and long-term estimated expectations of it's combined inputs. The output is the comparison of the short-term expectation relative to the long-term expectation.
Also included is a console program to facilitate training, testing, and analysis of individual Soevnns.
Please note: this is a work in progress.
Credits:
Soevnn - Copyright © 2021 Lilith Stanley
MessagePack.FSharpExtensions - Copyright © 2017 pocketberserker
MessagePack for C# - Copyright © 2017 Yoshifumi Kawai and contributors
lz4net - Copyright © 2013-2017 Milosz Krajewski
NUnit - Copyright © 2021 Charlie Poole, Rob Prouse