Enhancing Image Classification with Quantum-Classical Hybrid Convolutional Neural Networks (Spoke 6)

Anni Domenics Arias (Lutech, FAIR Spoke 6 - Symbiotic AI) presents "Enhancing Image Classification with Quantum-Classical Hybrid Convolutional Neural Networks".
This presentation is part of the Virtual Young Poster Session of the FAIR 2024 General Conference.
For more information: fondazione-fai...
A very relevant topic in Spoke 6 is the sustainability of artificial intelligence. Lutech spa, as an active company in the field of artificial intelligence and partner in the FAIR project, decided to pick up the challenge of sustainability. In this poster, we present a novel approach for image classification based on hybrid quantum- classical convolutional neural networks (QCCNNs). The developed prototype, inspired by the model discussed in [Henderson et al. (2019)], processes 2D input images with a variational quantum circuit functioning as a filter. Unlike classical convolutions requiring multiple filters for different output channels, the QCCNN employs a single quantum filter with arbitrarily adjustable parameters and can handle complexity thanks to quantum state superposition. Preliminary numerical tests on a synthetic “Tetris” dataset of 1000 3x3 gray-scale images showcased potential quantum benefits: noiseless QCCNNs outperformed classical CNNs in terms of classification accuracy and did so employing a fewer number of parameters. When a measure of simulated noise (comparable to realistic scenarios) is added to the system, QCCNNs are still able to match the performance of classical CNNs but using fewer parameters. We expect the QCCNN architecture to prove its success on real noisy intermediate-scale quantum devices as well, improving prediction accuracy while requiring fewer resources, thus improving sustainability of the whole AI system.