A linear photonic swap test circuit for quantum kernel estimation

Among supervised learning models, Support Vector Machine stands out as one of the most robust and efficient models for classifying data clusters. At the core of this method, a kernel function is employed to calculate the distance between different elements of the dataset, allowing for their classification. Since every kernel function can be expressed as a scalar product, we can estimate it using Quantum Mechanics, where probability amplitudes and scalar products are fundamental objects. The swap test, indeed, is a quantum algorithm capable of computing the scalar product of two arbitrary wavefunctions, potentially enabling a quantum speed-up. Here, we present an integrated photonic circuit designed to implement the swap test algorithm. Our approach relies solely on linear optical integrated components and qudits, represented by single photons from an attenuated laser beam propagating through a set of waveguides. By utilizing 2$^3$ spatial degrees of freedom for the qudits, we can configure all the necessary arrangements to set any two-qubits state and perform the swap test. This simplifies the requirements on the circuitry elements and eliminates the need for non-linearity, heralding, or post-selection to achieve multi-qubits gates. Our photonic swap test circuit successfully encodes two qubits and estimates their scalar product with a measured root mean square error smaller than 0.05. This result paves the way for the development of integrated photonic architectures capable of performing Quantum Machine Learning tasks with robust devices operating at room temperature.