The Differential Evolution algorithm (Storn & Price, 1997) was tested by Plagianakos and Vrahatis (Proc. of the IEEE Int. Joint Conf. on Neural Networks - IJCNN’2000) for solving the XOR problem by training a 2-2-1 MLP. Unfortunately, the authors did not include the source code. Does anyone know where can I get that code? Thanks in advance!
Colin, I'm not talking about "differential equations", but "differential evolution algorithm", an evolutionary strategy used for parameters optimization.
Please read again
Hi Erick,
I'm enclosing the MLP code for XOR problem using Matlab. This code only designed for a simple understanding purpose of MLP for XOR problem. It may Help you.
Thank you very much Anil!
If you know how train the multilayer perceptron by using the differential algorithm, please let me know...
Best regards!
Hi Erick,
I'm not much aware about how to train MLP with DA but individually GA, PSO etc code, I can provide (if needed) where you can apply your own fitness function, initial candidate population, stopping criteria and maximize, minimize the solution scope. .
Dear Anil, thank you again!
Let me explain in detail. I recently implemented a Spiking Neural Network and I would like to test whether it can be trained. I used Neuroids (a novel neuron-model introduced by me at EMBC 2012 - you may download the article from my profile page) to build this network, whose architecture is not similar to traditional topologies (feed-forward, recurrent, etc). I recently found out that the differential algorithm (DEA) introduced by Kenneth & Price (1997) is suitable to handle non-differentiable, non-linear networks (as the one I implemented), but I've found many difficulties in trying to adapt DEA to my problem.
Please let me know if you want further details regarding this issue... and thank you again!
Best!
I did it!
Here is the matlab code for solving the XOR problem using a Multilayer Perceptron (2-2-1) trained by the differential algorithm (see the attachment)
Thanks anyway!
- Badges
- Science topic