Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
QI: Fachverband Quanteninformation
QI 16: Quantum Software
QI 16.8: Vortrag
Donnerstag, 12. März 2026, 12:00–12:15, BEY/0245
Machine Learning methods for Entanglement Detection in High Dimensional Systems — •Yasmin Bougammoura1, Martin Plávala1, Fabio Anselmi2, and Fabio Benatti2 — 1Institute of Theoretical Physics - Leibniz University Hannover, Germany — 2University of Trieste, Italy
A universal approach to entanglement detection can be cast as a semidefinite program (SDP), but this formulation becomes computationally inefficient as the dimension of the system grows. Since entanglement detection for systems of dimension d ≥ 6 is already an NP-hard task, this motivates the search for practical numerical strategies. Although current machine learning models reach high accuracy values, they do not provide robust and verifiable separability criteria. We propose two simple approaches: i. using automatic differentiation to approximate a given quantum state, and ii. training an artificial neural network to construct a certifiable entanglement witness W for a given entangled quantum state. The former efficiently approximates separable states; the latter achieves an improvement of two orders of magnitude in the dimension of the optimisation problem, which increases exponentially with the number of qubits. In terms of computational resources, the neural network model runs on a local machine with 16 GB RAM for a system of 8 qubits, whereas the SDP formulation is limited to a system of only 4 qubits.
Keywords: entanglement; automatic differentiation; neural networks; numerical optimisation; high dimensional systems