QSDL (101202109)

  https://cordis.europa.eu/project/id/101202109

  Horizon Europe (2021-2027)

  Quantum-secure multi-party deep learning

  MSCA Postdoctoral Fellowships 2024 (HORIZON-MSCA-2024-PF-01-01)

  revolutions

  2025-06-01 Start Date (YY-MM-DD)

  2028-05-31 End Date (YY-MM-DD)

  € 0 Total Cost

  Description

Deep Neural Networks (DNNs) have sparked a revolution across multiple domains. However, the escalating computational demands of advanced DNNs have led to the need for high-power consumption accelerators, which hinders their deployment on low-power devices. To enable advanced DNNs on such devices, the industry has resorted to offloading computationally intensive DNN inference to cloud servers. Nevertheless, this offloading architecture introduces vulnerabilities that compromise data security, presenting a pressing challenge, particularly in AI applications where data privacy is essential. In response, I propose a pioneering approach that harnesses the quantum nature of light to establish a practical paradigm for information-theoretically secure deep learning, where clients safeguard their sensitive data, and servers protect their proprietary AI models. This project delves into information science, aiming to explore theoretically and experimentally whether quantum science can offer practical advantages in privacy-preserving inner products against limited adversaries. The innovative essence of this proposal lies in the demonstration of quantum private deep learning in Prof. Englund's lab at MIT, utilizing an optical network. The experimental validation of the protocol with real-life healthcare devices in collaboration with Prof. Katz (HUJI) will unlock a new era of machine learning applications, including the secure utilization of AI in healthcare (e.g., IBM's Watson, Aidoc’s aiOS) and AI assistance (e.g., ChatGPT, Bard). This work may also reshape regulations like the EU General Data Protection Regulation.

  Complicit Organisations

1 Israeli organisation participates in QSDL.