LEP-CNN: a lightweight edge-device-assisted privacy-preserving CNN inference scheme for IoT

LEP-CNN is a lightweight edge-device-assisted privacy-preserving CNN inference scheme for IoT that lets IoT devices securely offload more than 99% of CNN computation. The scheme exploits the linearity of the most time-consuming CNN operations to precompute decryption offline, so the edge device performs inference on encrypted data as efficiently as on plaintext. AlexNet experiments show inference speedup of more than 35×, an integrity check success rate of more than 99% in identifying incorrect results, and better performance than CryptoNets-based homomorphic encryption schemes in time-sensitive scenarios.