Supervisor: Alexandru Gheorghiu (California Institute of Technology)
Abstract: Noisy intermediate-scale quantum (NISQ) devices face challenges in achieving high-fidelity computations due to hardware-specific noise. As a basis for noise mitigation, we develop a convolutional neural network noise model to estimate the difference in noise between a given pair of equivalent quantum circuits. On a classically simulated dataset of 1.6 million pairs of quantum circuits with a simplified noise model calibrated to IBM Q hardware, the deep learning approach shows a significant improvement in noise prediction over linear gate count models. A greedy peephole optimization procedure is proposed to minimize noise using the deep learning model as an objective function, showing further improvement in noise mitigation compared to commonly used gate count minimization heuristics.
ACM-SRC Semi-Finalist: yes
Poster Summary: PDF
Back to Poster Archive Listing