fluffything

103 ms·

1. 931.▲
   by fluffything 8y ago

   Do you have a proof that the neural network approximates the function within some epsilon for all possible inputs within some range ?
2. 932.▲
   by fluffything 8y ago

   People routinely use lagrange polynomials, fourier approximations, etc. with 10^10 or more coefficients and many more data-points. That's often orders of magnitude larger than what training most neural nets for AI need.
3. 933.▲
   by fluffything 8y ago

   So can any lagrange polynomial, Fourier series, ... The real question is whether the approximation is a good one: - can you prove error bounds ? - can you bound the maximum error? - is it efficient ? (low storage, low computational effort)