Quantum Machine Learning
Research overview
We study how quantum computers can be used for machine learning and optimization. Our work includes designing algorithms for data classification and combinatorial optimization with quantum circuits, as well as using mathematics to understand the advantages and limits of quantum computation.
The Quantum Machine Learning (QML) team studies variational quantum algorithms, quantum machine learning, and combinatorial optimization, primarily through theory and numerical simulations. Much of our current work focuses on optimization, including methods that draw on compressed sensing. We also study algorithms that can run with a small number of qubits or shallow quantum circuits.
Using quantum computers for combinatorial optimization
Combinatorial optimization involves finding a good combination that meets given conditions among many possible candidates. These problems become harder to compute as their size increases. We study how methods such as the Quantum Approximate Optimization Algorithm (QAOA) can be applied to them. We also develop optimization algorithms that use ideas from compressed sensing to find good solutions with fewer evaluations.
Examples of our research: compressed sensing with QAOA, and combinatorial optimization using Monte-Carlo Compressive Optimization
Using quantum circuits as learning models
Quantum circuits can perform machine learning tasks such as classification by taking data as input and learning circuit parameters. We study the design and training of quantum classifiers, including ways to learn with limited quantum resources and fewer measurements. In collaborative experiments with photonic quantum circuits, we also test performance on real devices.
Examples of our research: Fully Quantum Classifier, and a single-qubit classifier trained with a small number of samples
What can small quantum circuits compute?
Current quantum computers are limited in the number of qubits and circuit depth they can use. We investigate what can be computed with shallow circuits or a small number of qubits through both theory and experiments. Our work includes comparing quantum and classical processors under the same conditions and studying how quantum computation can be used for tensor-network approximations.
Examples of our research: demonstrating quantum advantage in a Sequential Processor, and tensor-network approximations with low bond dimension
How we conduct research
The Quantum Machine Learning team works mainly with mathematical models and numerical simulations. We formulate research questions as learning or optimization problems, then design quantum circuits and algorithms. We evaluate their performance through simulations and use comparisons with classical methods and theoretical analysis to determine the conditions under which they work. Some projects also involve experimental tests with collaborators using photonic quantum circuits.