Hybrid quantum-classical multiple kernel methods for machine learning

Quantum computers are well-suited to a subset of machine learning tasks known as supervised learning
(SL). In SL, a model is trained on labelled data to generate correct labels for novel data thereafter. Robust
classifiers can be achieved in SL by using the “kernel trick”, which provides a more expressive metric from a
transformed point of view. Strong theoretical and experimental results suggest an advantage in outsourcing
the kernel computation to a quantum processor; in an approach compatible with modular quantum devices.
The goal of this project is to determine which combinations of quantum and classical kernels are suited to
which types of data, and how the combined approach can improve upon single quantum kernels. Because
the near future likely entails noisy intermediate-scale quantum computers, the hybrid approach is among the
most viable for near-term quantum machine learning. Therefore, advances in this method will directly benefit
the partner organization, whose stated purpose is to develop software for the quantum era.

Faculty Supervisor:

Juan Felipe Carrasquilla

Student:

Partner:

Agnostiq

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Waterloo

Program: