MIT News September 26, 2024
Secure multiparty computations are typically offloaded to cloud computing servers, leading to vulnerabilities that can compromise the security of the clients’ data. A team of researchers in the US (MIT, industry) introduced a linear algebra engine that leveraged the quantum nature of light for information theoretically secure multiparty computation using only conventional telecommunication components. They applied this linear algebra engine to deep learning and derived rigorous upper bounds on the information leakage of both the deep neural network weights and the client’s data and obtained test accuracies exceeding 96% while leaking less than 0.1 bits per weight symbol and 0.01 bits per data symbol. This weight leakage was an order of magnitude below the minimum bit precision required for accurate deep learning using state-of-the-art quantization techniques. According to the researchers their work lays the foundation for practical quantum-secure computation and unlocks secure cloud deep learning as a field… read more. Open Access TECHNICAL ARTICLE
Optical implementation. Credit: arXiv (2024)