Comment by wookietrader

Wrong.

Neural networks are affinity invariant. You can rotate, skew, translate the input data however you want, the optimum stays the same. Same for SVMs.