Lets say I have 3 features a,b,c and lets say I am sure they vary between 0 to 100.

Instead of feeding my neural network 3 features what happens if I create a unique number from this 3 numbers something like 10000*a + 100*b + c = d and feed my neural network with d after some feature scaling?

I can see convolution is very common(which is different because it is not unique and more like a mean) that does not suits me because a,b,c are time series and their order matters.

Is there any advantage or disadvantage of combining multiple inputs into a single value in Neural networks?

  • $\begingroup$ Convolutions are not like a mean, but rather learn translation invariant patterns. If altering the features like you say is the best way, CNNs would learn that. $\endgroup$ – bonfab Jun 1 at 19:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.