Southwest Quantum Information and Technology

Entanglement detection using neural networks

Presenting Author: Diego Alberto Olvera Millán, Universidad Nacional Autónoma de México
Contributing Author(s): Pablo Barberis Blostein

Entanglement is an important resource for quantum technologies, but its detection and classification cannot be performed efficiently across different kinds of quantum states. In particular, quantum mixed states require computationally demanding methods, such as those of convex roof constructions. In this work, I train an artificial neural network(ANN) to perform the classification between entangled and separable two qubit states, using expected values of products of Pauli matrices as the entries of the feature vector, x. The training is performed using only random pure states sampled from the invariant Haar measure. It is found that, using the 15 linearly independent products of Pauli matrices, an accuracy of 98% is achieved for states drawn from the same distribution, and the accuracy for states drawn from the Bures distribution can reach up to 80% (after applying regularization to the model). Using 4 non orthogonal products of the Pauli matrices, an accuracy of 91% is achieved for states sampled from the Haar distribution, and, when dealing with states sampled from the Bures distribution with purity and concurrence higher than 0.7, an accuracy of up to 84% is achieved.

(Session 5 : Thursday from 12:00pm-2:00 pm)