Quantum-Enhanced Pattern Recognition

The challenge of pattern recognition is to invoke a strategy that can accurately extract features of a dataset and classify its samples. In realistic scenarios, this dataset may be a physical system from which we want to retrieve information, such as in the readout of optical classical memories. The theoretical and experimental development of quantum reading has demonstrated that the readout of optical memories can be significantly enhanced through the use of quantum resources (namely, entangled input states) over that of the best classical strategies. However, the practicality of this quantum advantage hinges upon the scalability of quantum reading, and up to now its experimental demonstration has been limited to individual cells. In this work, we demonstrate quantum advantage at a fixed resource, namely, at fixed mean probe energy, in the multicell problem of pattern recognition. Through experimental realizations of digits from the Modified National Institute of Standards and Technology (MNIST) handwritten digit dataset, and the application of advanced classical postprocessing, we report the use of entangled probe states and photon counting to achieve quantum advantage in classification error over that achieved with classical resources, confirming that the advantage gained through quantum sensors can be sustained throughout pattern recognition and complex postprocessing. This motivates future developments of quantum-enhanced pattern recognition of bosonic loss within complex domains.