QUANTUM PHOTONIC TRANSFORMER (QPT)

Abstract

In this thesis, I propose a classic-quantum hybrid Machine Learning model, Quantum Photonic Transformer (QPT). The model follows the transformer architecture that changed the landscape of the Natural Language Processing (NLP) since its invention. However, instead of the the dot-product and Softmax in the attention weight calculation, QPT uses simple quantum photonic circuitry to obtain the query-key similarity in the form of a Gaussian Radial Basis Function (RBF). By using the 2-dimensional noisy (spiral) classification as a toy problem, I show that the suggested attention mechanism in the classical regime (shots = ∞) improves the performance of the original self-attention.