New QEC Method Boosts QCNN Reliability with Low Overhead

Quantum Convolutional Neural Networks (QCNNs) hold promise for accelerating classification tasks by combining quantum computing with classical CNNs. However, the high noise levels inherent in current quantum hardware severely hinder their practical execution, often preventing convergence. While robust error correction methods like the surface code exist, their qubit cost is prohibitively large for many applications. This research introduces a novel, low-overhead quantum error-correction (QEC) technique specifically designed for QCNNs. The method utilizes distance-4 bivariate bicycle (BB) codes, which are notable for their high error threshold, constant encoding rate, and linear code distance. Simulations with realistic hardware noise sources demonstrated that an unprotected 4-qubit QCNN failed to converge and exhibited poor learning. By applying the proposed BB QEC technique, the researchers validated that their method enables QCNNs to converge and achieve better learning rates. This represents a significant step towards making practical QCNNs feasible on noisy intermediate-scale quantum (NISQ) devices, overcoming a major hurdle in quantum machine learning.