Unlocking Privacy and Accountability: Zero-Knowledge Proofs in Machine Learning