Method for anomaly detection within a dataset, specifically designed for being implemented on a quantum annealing device

Abstract

A method for anomaly detection within a dataset X of N data points is implemented on a quantum annealing device including a plurality M of qubits, with M≥N. A cost function to be minimized is defined as: where dis a distance between a point x∈X and a centroid of the dataset distribution, dis a distance between two data points x, x∈X with i≠j, α∈is a weighting parameter, and x, x={0, 1}. If a qubit can interact with a maximum determined number of different qubits, applying the cost function q(x, α) to the plurality of qubits comprises limiting a number of quadratic terms for each variable xto a value k smaller than or equal to the maximum determined number of interactions between qubits within the quantum annealing device. An assignment of binary values to the variables in the dataset that minimizes the cost function is calculated, and the k variables in the dataset associated with qubits of value 1 are detected as anomalies.