Machine learning-based device-independent certification of quantum networks

Description

The detection of non-classical features, along with the possibility of carrying out optimizations over the set of quantum behaviors, has a great relevance in the field of quantum technology. In particular this task becomes pivotal, when the goal is assessing the reliability of quantum hardware in a quantitative way. At the state of art, the most widely employed techniques to approximate the set of quantum correlations (which is a hard computational task) rely on semidefinite programming optimization (SDP). However, such a tool has two intrinsic limitations. Firstly, it can be applied only to linear objective functions and enforce linear constraints, therefore excluding quantum networks featuring independent hidden variables. Moreover, such optimizations become computationally unfeasible when the complexity and size of the system grow. To circumvent this issue, we introduce an artificial neural network-based strategy, that allows to carry out numerical optimizations over supersets of the quantum
set. This method has two main advantages: on one side, it can be applied to nonlinear optimizations constraints and objective functions. This opens the possibility of studying quantum networks featuring independent sources or to optimize nonlinear entanglement witnesses. On the other hand, it requires less computational resources than the aforementioned SDP-based techniques, allowing a better scalability and promising to be a useful tool to explore more complex scenarios.

Period	15 Apr 2023
Event title	Machine Learning and (Quantum) Physics Workshop 2023: Physics and Machine Learning
Event type	Seminar/Workshop
Location	Obergurgl, AustriaShow on map
Degree of Recognition	National