 |
Center for Nonlinear Studies |
will begin by demonstrating that the answer to the first question in the title is yes [1], in principle. I will then discuss ifthe quantum advantage of quantum machine learning can be exploited in practice. To discuss how to build optimalquantum machine learning models, I will describe our recent work [2-3] on applications of classical Bayesian machinelearning for quantum predictions by extrapolation. In particular, I will show that machine learning models canbe designed to learn from observables in one quantum phase and make predictions of phase transitions as well assystem properties in other phases. I will also show that machine learning models can be designed to learn from datain a lower-dimensional Hilbert space to make predictions for quantum systems living in higher-dimensional Hilbertspaces. I will then demonstrate that the same Bayesian algorithm can be extended to design gate sequences of aquantum computer that produce performant quantum kernels for data-starved classification tasks [4].
[1] J. Jager and R. V. Krems, Universal expressiveness of variational quantum classifiers and quantum kernels for supportvector machines, Nature Communications 14, 576 (2023).
[2] R. A. Vargas-Hernandez, J. Sous, M. Berciu, and R. V. Krems, Extrapolating quantum observables with machinelearning: Inferring multiple phase transitions from properties of a single phase, Physical Review Letters 121, 255702(2018).
[3] P. Kairon, J. Sous, M. Berciu and R. V. Krems, Extrapolation of polaron properties to low phonon frequencies byBayesian machine learning, Phys. Rev. B 109, 144523 (2024).
[4] E. Torabian and R. V. Krems, Compositional optimization of quantum circuits for quantum kernels of support vectormachines, Physical Review Research 5, 013211 (2023).
Bio: Roman Krems is a Professor of Chemistry and Distinguished University Scholar at the University of BritishColumbia. He is also a member of the computer science department at UBC and a principle investigator at theStewart Blusson Quantum Matter Institute. His work is at the intersection of quantum physics, machine learning andchemistry on problems of relevance to quantum materials and quantum technologies. He is particularly excitedabout applications of machine learning for solving complex quantum problems and applications of quantum hardwarefor machine learning. He is Fellow of the American Physical Society and Member of the College of the RoyalSociety of Canada.
Followed by Talks by Local LANL Scientists:
- 15:15 - 15:40 Frank Barrows, The Dynamics and Capacity of Neuromorphic Neural Networks and Learning by Mistakes
- 15:40 - 16:05 Avadh Saxena, Non-Hermitian Quantum Mechanics and its Physical Implications
- 16:05 - 16:30 Andrew Harter, Quantum Metrology with Non-Hermitian Rydberg Atoms