Is this basically the learning parity with noise problem, but instead of a fixed size vector, the noisy output is generated in by a PRNG with a pre-determined secret seed? How can this be used for authentication or encryption?