jussmith01 / neuralnetintaddition Goto Github PK
View Code? Open in Web Editor NEWMy first attempt at building a neural net of sigmoid neurons to compute n+1 given n for any integer n. This is a self project to help with my understanding of how neural nets work in practice. Derivatives are calculated using a simple gradient descent algorithm. This program builds a training set of 32-bit integers, converts them to binary, converts the binary digits to double precision which it uses as input to the neural net. The output is expected to be a double point number between 0 and 1, and if the number is < 0.5 it is converted to a 0, if > 0.5 it is converted to 1 then the binary digits can be converted back to a 32 bit int. For the training set I generate a set of random 32 bit integers. Summer 2015.