Researchers at IBM Quantum Investigate Potential of Quantum Kernel Methods
(Phys.org) Over the past few years, some scientists specialized in quantum algorithms have thus been exploring the potential of quantum kernel methods, which were first introduced by Havlicek and his colleagues at IBM.
Researchers at IBM Quantum have recently carried out a study further investigating the potential of quantum kernel methods. Their paper, published in Nature Physics, demonstrates that these methods could provide a robust quantum speed up over conventional kernel methods.
“Despite the popularity of quantum kernel methods, a fundamental question remained unanswered: Can quantum computers employ kernel methods to provide a provable advantage over classical learning algorithms?” Srinivasan Arunachalam, one of the researchers who carried out the study, told Phys.org. “Understanding this question was the starting point of our work. In this Nature Physics paper, along with my collaborators Yunchao Liu and Kristan Temme, we resolved this question in the affirmative.”
The recent work by this team of researchers provides a confirmation that quantum kernel methods could help to complete classification tasks faster and more efficiently. In their future studies, Arunachalam and his colleagues plan to investigate the potential of using these algorithms to tackle real world classification problems.
“The classification problem that we used to prove this advantage is artificially constructed to provide a theoretical underpinning for the usefulness of quantum kernels,” Arunachalam said. “There is room to obtain further quantum speedups using quantum kernel methods for other (hopefully) practically relevant problems. We believe our result is interesting because it provides us with a direction to look for more learning problems that can benefit from kernel methods. In our future work we hope to understand how generalizable the structure of our classification problem is and if there are further speedups obtainable using similar structures.”