Researchers at IFISC (UIB-CSIC) have proposed a novel method for pattern classification using a single nonlinear quantum oscillator instead of a lattice of spins (bits). The study, published in the prestigious journal Physical Review Letters, focuses on the implementation of the associative memory method and classification algorithms using AI.
Until now, pattern classification tasks were performed by implementing bit or spin networks in which these elements interact with each other classically or quantum, respectively. However, these methods present problems in terms of the ability to recognize a large number of different patterns, as this capability is strongly limited by the size of the system. The proposed method published by the IFISC researchers is based on replacing the spin lattice by a single quantum oscillator periodically forced and with nonlinear dissipation terms. These nonlinear terms allow the system to have a metastable phase in which the patterns to be identified can be encoded in coherent states of light that act as attractors, to which the system has a tendency to direct its dynamics
The energy levels of the quantum oscillator can encode the different patterns that the method will classify. This is an important advantage since it works with a single system, as opposed to a network of bits or spins in which it is necessary to build a lattice of interacting elements. In addition, the study demonstrates that with a smaller system size more patterns can be discerned, overcoming the theoretical limit presented by quantum bit lattices.
This novel method of pattern classification is performed in the transient phase of the system, allowing results to be obtained more quickly and, therefore, preventing quantum information from being lost. The study represents a significant advance in the field of pattern recognition and classification using quantum physics.
Artificial neural networks are brain-inspired computational systems that can solve and model numerous types of tasks, from pattern and speech recognition to big data analysis. The proposal opens up new possibilities in implementing time-stable solutions using quantum oscillators for a variety of problems, such as pattern recognition.
Labay-Mora, A., Zambrini, R., & Giorgi, G. L. (2023). Quantum associative memory with a single driven dissipative nonlinear oscillator. Physical Review Letters, 130(19). https://doi.org/10.1103/physrevlett.130.190602