NeuronDotNet - Neural Networks in C# / News: Recent posts

NeuronDotNet is a C# neural network engine. It provides an interface for advanced AI programmers to design various types of artificial neural networks and use them. It is compatible with .NET framework 2.0

NeuronDotNet 3.0 features

* Support for neural networks with any acyclic structure of layers
* One-One and Complete connectors are supported
* Backpropagation networks and Kohonen SOMs are supported
* Learning Rate changes from its initial value to a final value using a pluggable function (Linear, Logarithmic and Hyperbolic functions are built in)
* Neural network initialization modules are pluggable (Random, Constant, NguyenWidrow and Normalized Random Functions are built in)
* Custom activation funtions used in backpropagation networks are pluggable (Sigmoid, sine, tanh, logarithmic and linear functions are built in)
* For a Kohonen Layer, Neighborhood functions are pluggable (Gaussian function and Mexican Hat functions are built in)
* Various events are exposed which allow users to analyse how a network learns
* Kohonen layers are planar in shape. However, we can have circular rows and/or columns which make them attain the shape of a cylindrical surface or a toroidal surface.
* Hexagonal and Rectangular Kohonen lattice topologies are supported
* Training set has been defined to support Batch Training
* API to add custom network architectures and learning algorithms
* Layers, connectors, networks and training sets implement ISerializable interface

NeuronDotNet is a C# back propagation neural network engine. It provides an interface for advanced AI programmers to design various types of artificial neural networks and use them. It is compatible with .NET framework 2.0

NeuronDotNet Core 2.0 is completely redesigned from scratch and supports following features

(i) Neural networks with any structure
(ii) Different types of Activation Functions
(iii)Enhanced Back Propagation Algorithm (using momentum term, weight Decay and jitter).
(iv) Simulated Annealing (to accelerate the training process)
(v) Complete and One-One connections between layers... read more