Combining Machine Learning and Homomorphic Encryption in the Apple Ecosystem (machinelearning.apple.com)

Apple has announced a significant advancement in its commitment to user privacy by integrating homomorphic encryption (HE) within its ecosystem, particularly to enhance on-device machine learning (ML) functionalities. This innovative approach allows for secure server lookups while keeping users' data confidential. The HE technology enables computations on encrypted data, meaning that even during data processing, the server remains unaware of the original requests. Apple's implementation leverages the Brakerski-Fan-Vercauteren (BFV) HE scheme, which supports vital operations for ML, such as dot products, while ensuring post-quantum security standard at 128 bits. The use of HE, along with complementary privacy-preserving methods like private information retrieval (PIR) and private nearest neighbor search (PNNS), allows Apple to deliver features like Enhanced Visual Search without compromising user privacy. For instance, users can search their photo libraries for landmarks while their data remains secure, as queries are anonymized and encrypted. By open-sourcing the swift-homomorphic-encryption library, Apple aims to empower developers to adopt similar privacy-first practices in their applications. This development is significant for the AI/ML community as it sets a new standard for privacy in machine learning applications, balancing computational efficacy with strong data protection.