Verifiable Computation for Approximate Homomorphic Encryption Secures Private AI
New HE-IOP primitive solves the integrity problem for approximate homomorphic encryption, enabling verifiable, private, outsourced computation for AI models.