Overview

Machine learning empowers computers to solve complex tasks such as pattern identification and strategy optimization with applications in, e.g. financial trading, fraud detection, medical diagnosis, and self-driving vehicles. The required computing power is, however, pushing existing computational resources to their limits, restraining their further advancement. In QFreC, I target the realization of photonic frequency-based quantum co-processors, specifically tailor-made to solve machine learning problems with capabilities commensurate with today’s high-power, yet energy-efficient processing needs. In particular, I will use a high-dimensional photonic quantum frequency comb approach, where photons have hundreds to thousands of discrete and equidistantly spaced frequency modes, giving access to large, scalable information capacity.

For implementing quantum-accelerated machine learning tasks such as the classification of classical or quantum data, I will follow i) the exploration of quantum photonic frequency-domain processing with the adaptation of qubit learning concepts (vector-based and neural network-based approaches) to high-dimensional quantum representations, i.e. quDits, ii) the realization of efficiency-enhanced and novel integrated quantum frequency comb systems with quantum resources that allow real-world applications using highly nonlinear on-chip platforms, and iii) the development of reconfigurable, fast, and broadband experimental control schemes using, e.g. quadrature amplitude modulation formats and nonlinear optical processes. To enable stable, compact, cost- and energy-efficient quantum processing devices, the QFreC project will build on the advances of the well-developed telecommunications infrastructure and the photonic chip fabrication industry.

QFreC merges photonic quantum frequency-domain circuits with quantum machine learning, enabling large-scale controllable quantum resources for the exploration of quantum-enhanced machine learning.