DynamicDataAndExamples
Class Library for creating and using unstructured neural networks
This framework is a work in progress. When it is finished it will be a framework for compiling a computer language to be used for data processing and the creation of neural networks.
In its current state it can be used as a C# Class Library for the creation of unstructured neural networks.
Unstructured Neural Networks
I demonstrate in this project that you do not need structured sets of nodes and backpropagation to achieve a functioning neural network. I am hoping that this proves to be true and as of right now it does appear to be. However, I am limited in my time that I can work on this project so I am giving away the source code(MIT Licence)
What are potential advantages of this system?
Youtube video explaining the network: https://youtu.be/UZJQNL0Wt3A
Has this been done before?
Not to my knowledge but if you know of a system like this please let me know.
Why Get Rid of Back Propagation?
Because backpropagation is the more time consuming and difficult element of neural networks. It requires an enormous amount of tweaking to get it correctly assigning blame to individual nodes. You have to either understand precisely what is needing to be achieved or you go through an enormous amount of iterations, trying different things. Both options are prohibitable expensive. Even if you were to create a neural net to handle backpropagation it would be limited by its own backpropagation. Not to mention backpropagation is also resource intensive.
How The FUCK Did You Eliminate Backpropagation!?!?
Ok so… Your just going to have to watch the youtube video I linked above. It's basically chaos theory written in code. In other words, the system that wins wins.
I will add more details later but for now I’m not sure what else to write.