Artificial Neural Network Component for Grasshopper

- Implementation of NeuronDotNet in a computational design environment -



Backpropagation Network in Grasshopper
from Benjamin Felbrich on Vimeo.

2-dimensional Self-Organizing Map in Grasshopper
from Benjamin Felbrich on Vimeo.

n-Dimensional Self-Organizing Map in Grasshopper
from Benjamin Felbrich on Vimeo.


The mathematical and computational model of artificial neural networks - the idea of passing numbers through interconnected nodes and adjusting how strong these "signals" are pursued over the net topology by assigning node weights, edge weights and neighborhood functions - is more than 70 years old.
However, since Warren McCulloch and Walter Pitts first published this idea of a "threshold logic" in 1943, the Minsky-Papert perceptron was introduced in 1969 and since backpropagation was firstly applied for error elimination in 1975 by Werbos, a lot of advancements took place.
Today, many sophisticated engines for basic artificial neural networks, Kohonen networks and deep learning networks are available.
Yet they've never been utilized as a tool to design and optimize buildings.
Artificial neural networks bear the ability to arrange and structure inputs of arbitrary dimensions.
In a 3D space they can be used to fit meshes and grids through odd point clouds. In this manner they can organize and map a given topology into these points.
In higher dimensions artificial neural networks are useful to structure and cluster data of any kind. Even pattern recoginition of fuzzy inputs are possible.
The idea behind this project was to implement an existing engine for artificial neural networks, called NeuronDotNet3.0 developed by Vijeth Dinesha in the Grasshopper environment.
Why "Crow"? Because crows are smart.

Crow is based on the C# library NeuronDotNet, developed by Vijeth Dinesha and the NeuronDotNet team. Copyright 2008 Vijeth D.