Pasqal, developers of neutral atom-based quantum technology, today announced the publication of a scientific paper in the peer-reviewed APS Physics journal Physical Review A presenting a new machine learning protocol for measuring the similarity between graph-structured data on quantum computers.
Titled "Quantum evolution kernel: Machine learning on graphs with programmable arrays of qubits," the study details how a Quantum Evolution Kernel (QEK) can serve as a more versatile and scalable procedure for building graph kernels and analyzing graph-structured data on quantum devices as compared to classical computers. Benchmarking expected performances on a neutral-atom quantum computer, researchers found that QEK is stable against detection error and on par with state-of-the-art graph kernels on classical systems in terms of accuracy.
Graph kernels are computer functions that measure the similarity of pairs of graphs in machine learning applications but their complexity have posed a challenge to classical computing systems. The study's participants believe the new QEK will enhance research in fields that utilize graph structures even when deployed on NISQ-era quantum devices, including chemistry, bioinformatics, computer vision, social network analysis and natural language processing.
"This new machine learning protocol for measuring the similarity between graphs leverages the core interactions that are at the heart of the quantum processor and can be readily implemented by researchers today," said Georges-Olivier Reymond, CEO of Pasqal. "The work of our quantum scientists also demonstrates the maturity of our neutral-atom platform in enabling this significant advancement."
Pasqal CTO Loïc Henriet served as the research lead for the paper working with other Pasqal researchers Louis-Paul Henry, Slimane Thabet and Constantin Dalyac.
The results in the paper were produced using Pasqal's open-source library Pulser and its embedded emulator. The high-performance computing center GENCI in Paris provided access to its supercomputers for the study.