alt Hacker NewsIt is applicable, but it is not useful.
Universal hashes use the input text as the set of coefficients.
This method requires additional preprocessing of the coefficients, before starting to evaluate the polynomial. That preprocessing would slow the hashing algorithm more than what is gained during evaluation.
This method is useful only when with a given polynomial, i.e. set of polynomial coefficients, you want to evaluate that polynomial many times, so the cost of the preprocessing is amortized.
However, this application is very important because most functions are approximated either with polynomials or with rational functions, so this method can accelerate the evaluation of all such approximated functions.
> This method requires additional preprocessing of the coefficients, before starting to evaluate the polynomial. That preprocessing would slow the hashing algorithm more than what is gained during evaluation.
There is no preprocessing at hash time in either use.
Universal hashing: the message words are the parameters of the chain, a_i and b_i in P_i = a_i + (b_i + y)(P_{i−1} + u), not coefficients of a target polynomial. Distinct messages give distinct polynomials, which is all a universal hash needs; the decoder never runs. Same as Bernstein's BRW.
k-independent hashing: the key should be a uniformly random monic polynomial of degree k. Our parameterisation is a bijection onto those polynomials, with the rational preprocessing as its inverse, so uniformly random gate constants give a uniformly random polynomial. You draw the ⌊k/2⌋+1 constants and evaluate; the coefficients are never computed. That is why the paper needs bijective rather than just injective constructions, and the Section 5 speedups are for the whole hash.
Preprocessing only appears when a fixed polynomial (a Taylor approximation, a secret-sharing polynomial) is evaluated at many points, and then it runs once.