I have weights and biases for my network model, training in MATLAB tool. I implemented the network model in MATLAB code and I compared the results with sim() function unfortunately I am getting wrong results.
Anyone implemented the network on microcontroller?
Can you specify your question a bit further? Your current statement is quite vague.
Some usefull additional details would be:
What microcontroller you are considering, what type of system you want to emulate, and if you want to implement the model in the controller for co-simulation, control or fault detection?
If you'd like to switch to a specific microcontroller, it could be usefull to investigate the programming languages and toolboxes available for that component.
Also, try to make sure the code, structure and model order is suitably chosen in Matlab (with acceptable fit in cross-validation), since otherwise you'll be implementing an useless model for your purpose.
Dear Anand,
I think these articles will help you to clarify your doubt. I will suggest you to go for MATLAB simulation platform rather than coding. You may get better result in this way. Otherwise LabView is the best tool available in present time for implementation of ANN logic in real-time circuits and networks.
http://www.iaeng.org/publication/WCECS2011/WCECS2011_pp183-188.pdf
http://cdn.intechopen.com/pdfs-wm/9274.pdf
Regards,
A.Kar
Explain more! Type of Neural Network that used (Recurrent or FeedForward), type of activation function and... ??
It should be considered that the most important feature of neural network is concurrent computing. But microcontrollers are sequential. So, for the good performance, the written code (for microcontroller) should be close to the parallel processing. I propose that to implement the Neural network on the FPGA or DSP.
Thank you all. I used fitting tool in MATLAB which is a feed forward tool and the activation function is 2./(1+exp(-2*n))-1, also generated by the tool.
- Badges
- Science topic