Implementation

Soevnn is a neural net consisting of a heterogeneous population of neurons connected in a directed graph. The neurons are not pre-organized into layers. Instead they are pre-organized into structures of groups of clusters. Each group consists of neurons of a neural type, with their own set of parameters. Neurons are connected to one another probabilistically. Additional, neurons are sorted into neural pathways which expedite the flow of information.

This is the 5th incarnation made from scratch, the first to be made using F#. Previous incarnations were made using C#. F# was chosen this time as first-class function support provides greater extensibility and immutable-by-default variables create stricter access control then C# access modifiers. F# is also better suited for flexible parallelism by virtue of the previous two benefits and a greater focus on the separation of functions and data.

While previous incarnations of Soevnn focused on visual demonstration via integration into a specialized art program, the current focus is analytical tools to study various instances of Soevnns. Additionally, unit tests are made use of to help ensure proper function and as a debug tool to more quickly locate the source of errors.