Crow - Artificial neural network component for Grasshopper

This clip shows a short demo of the plugin "Crow" for Grasshopper in Rhino. It employs artificial neural nets such as Kohonen networks and classic backpropagation networks to remap given topologies into odd point clouds (TSP and Seolf Organizing maps). Furthermore it can be used for petterns recongition.

For more infos about Crow visit: www.felbrich.com/projects/Crow/Crow.html

17 members like this

Get Embed Code

Tags:

Comment

You need to be a member of Grasshopper to add comments!

Comment by Benjamin Felbrich on July 22, 2014 at 2:01am: Hi Carlos,

yes I definitely want to, but I thought to "clean up" the code a bit first and maybe implement one or two more components.

Cheers,
Benjamin

Comment by Benjamin Felbrich on July 22, 2014 at 1:59am: Hey guys,

I'm glad there's so much interest in these methods. Please note, that this is just a preview, and the component is far away from being "complete".

Andy & Fred:
I've actually chosen NeuronDotNet since its very simple. However, its not being continued and there aren't many examples / explanations to be found on the web. So it's definitely worth, thinking about alternatives in the future.
I actually don't really know so far, what to do with it in a design context, but I was hoping that people could play with it and find some nice applications.

Dani:
Crow will definitely be continued, I actuallly just started working on it. When there's an actual useful release ready, I'll also provide detailed explanations.
The use of 3d points as data to be processed is indeed questionable, so for now we could use it to just play around with geometry and find shortest paths through point clouds and fit meshes. To do this, one wouldn't need further information, but the engine bears a lot more possibilities which I planning to implement.
(n-dimensional inputs to be structured and clustered, pattern recognition etc.)

So there's still a lot of work to do, I'll keep you updated and hope to be able to provide a proper release soon.

Cheers,
Benjamin

Comment by Carlos Pérez on July 22, 2014 at 1:44am: Hi Benjamin,

Very cool! Are you adding it to www.food4Rhino.com?

thanks,

Comment by Fred Becquelin on July 22, 2014 at 1:17am: I was planning to use Aforge.NET framework some day for ANN, since I already used it for fuzzy logic (see my page) it made sense. AForge has a lot of awesome stuff inside :)

However I never figured how to use them in a design process.

Wait, why don't we quit design? Better start forecasting stock prices whith this!

Comment by Daniel González Abalde on July 21, 2014 at 8:28pm: Thanks for coding this in GH, biological methods for programming is powerful for designers and researchers. I admit my total ignorance on the matter but I have a lot of interest, I searched for information on kohonen networks, but I do not understand the concept of your component. I have a few questions, if you would be so kind to answer...

1. A prior/specific knowledge is required to use the component?

2. Crow component will continue to be developed?

3. What is the logic/meaning of the parameters of the net-map to use points (as data)?

4. Could u provide a sample ghfile (or pdf..) explained in detail?

5. Could u give an example of use on self-organized networks in design or architecture?

6. The result serves as a search for a goal or also serves as a design tool in itself?

Thank you.

Comment by Andy Payne on July 21, 2014 at 2:53pm: Nice. I see you're using an implementation of the NeuronDotNet library. I'm wondering if you've compared other libraries for speed/stability. I've used the Fast Artificial Neural Network (FANN) in the past and it worked quite well. Glad to see this coming into the Grasshopper environment!