zkTaylor: Zero Knowledge Proofs for Machine Learning via Taylor Series Transformation

Abstract

In order to enable more types of machine learning models to use zero-knowledge proofs to enhance their computational verifiability, this study proposes a zero-knowledge machine learning conversion method based on the Taylor series. Firstly, a polynomial expansion of structures with transcendental functions in ordinary machine learning models is performed using Taylor's formula. The corresponding arithmetic circuit descriptions are written in ZKP based on the converted model structures. Finally, the proof body is generated, which allows the verifier to verify the correctness of the results quickly. The basic experimental idea is also given, and the scheme's feasibility is verified, which can be done to provide a verification path for the model without seriously affecting its accuracy.