Quantum-secure multiparty deep learning: theory and experiments

The demand for cloud-based deep learning has intensified the need for practical and readily deployable secure multiparty computation. In this talk, I will present a linear algebra engine that leverages the quantum nature of light for provably secure computation using telecom components. Applied to deep learning, our approach achieves near-digital accuracy on handwritten digit classification benchmarks while guaranteeing information-theoretic bounds on data leakage [Sulimany et al., arXiv:2408.05629 (2024)]. Last, I will present our experimental demonstrations using rectification in thin-film lithium niobate photonic crystal cavities and waveguides [Sulimany et al., (2025). Accepted to CLEO].