Since you sound like an engineering type, brush up on the chain rule, and understand the math of backpropagation, THEN go see the YouTube video, Next Generation Neural Networks to see what the newest extension of backprogation is, as of November 2007; it rocks; fundamental change to modelling with neural nets
But it always helps to have solid mathematical foundations, THEN you can play with research ideas instead of simulators made by others.
This was a project for one of my classes, there is a slide show that explains things. The goal was to use a cheap (integer only) uC with DSP instructions that could accelerate a neural network. I didn't have time to implement XOR. But I will be taking a class on ANN this semester so if you have any suggestions pass them on.
Well, it depends how complete you want to be.
Since you sound like an engineering type, brush up on the chain rule, and understand the math of backpropagation, THEN go see the YouTube video, Next Generation Neural Networks to see what the newest extension of backprogation is, as of November 2007; it rocks; fundamental change to modelling with neural nets
But it always helps to have solid mathematical foundations, THEN you can play with research ideas instead of simulators made by others.
nightowl8936 4 years ago
How about you get it to learn the classic XOR?
Also, binary valued inputs are simpler than continuosly valued inputs.
What learning algorithm are you using? The old Perceptrons could learn basic Boolean stuff of the type you are doing, EXCEPT XOR.
Also, if its PIC... is the neural net programmed in software? Or is it using hardare of some kind on chip?
nightowl8936 4 years ago
This was a project for one of my classes, there is a slide show that explains things. The goal was to use a cheap (integer only) uC with DSP instructions that could accelerate a neural network. I didn't have time to implement XOR. But I will be taking a class on ANN this semester so if you have any suggestions pass them on.
inhuman4 4 years ago