Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session S69: Quantum Machine LearningInvited
|
Hide Abstracts |
Sponsoring Units: DQI Chair: Alejandro Perdomo-Ortiz, Zapata Computing Inc Room: Room 421 |
Thursday, March 9, 2023 8:00AM - 8:36AM |
S69.00001: Geometric Quantum Machine Learning Invited Speaker: Marco Cerezo Recognizing the underlying symmetries in a given dataset has recently played a fundamental role in classical machine learning, leading to the burgeoning field of geometric deep learning. Some of the ideas of geometric deep learning have been imported into the field of quantum machine learning, leading to a novel field that has been termed geometric quantum machine learning (GQML). In this talk, we will review the basic concepts of GQML. We will begin with a comprehensive introduction to the necessary tools from representation theory to understand and manipulate symmetries in the dataset. Then, we will show how to create models encoding the symmetries of the learning task. This is materialized through the usage of equivariant neural networks whose action commutes with that of the symmetry. Finally, we will show how equivariant quantum neural networks can solve many of the critical issues in variational quantum machine learning. In particular, we will prove that permutation-equivariant architectures do not suffer from barren plateaus, quickly reach overparametrization, and can generalize well from small amounts of data. |
Thursday, March 9, 2023 8:36AM - 9:12AM |
S69.00002: Scaling (generative) quantum machine learning models Invited Speaker: Christa Zoufal With the advancement of quantum technology, researchers aim to understand if and how quantum algorithms could have advantages compared to their classical counterparts, e.g., in the context of machine learning. The investigation of possible benefits of quantum compared to classical machine learning models requires thorough theoretical as well as empirical studies. In this context, various quantum machine learning algorithms have been studied that are based on short-depth, parameterized quantum circuits, which are well suited for execution on near-term quantum hardware. These models are promising candidates for a set of near-term empirical studies targeted to understand the applicability of quantum machine learning. However, as shown by a variety of research [1–4] training these models can become challenging, especially at increasing scale. In this talk, we discuss a set of challenges that particularly generative quantum machine learning [5–7] has to face and demonstrate potential remedies thereof in experiments with non-trivial qubit numbers. The illustration of a set of empirical results will additionally accentuate the presented problems and approaches. |
Thursday, March 9, 2023 9:12AM - 9:48AM |
S69.00003: TBD Invited Speaker: Sukin Sim
|
Thursday, March 9, 2023 9:48AM - 10:24AM |
S69.00004: Potential and limitations of probabilistic modelling with quantum circuits Invited Speaker: Ryan Sweke Learning representations of probability distributions is a fundamental task in machine learning to which quantum learning algorithms seem potentially well suited. In this talk I will describe some recent results, which contribute to our understanding of the extent to which different types of quantum probabilistic modelling algorithms may or may not offer concrete advantages over classical algorithms. Firstly, I will show that there exists a finely tuned probabilistic modelling task which is provably hard for classical algorithms, but efficiently solvable by a special-purpose quantum algorithm running on a fault-tolerant quantum computer. While this seems like a promising start, ideally one would like to show similar quantum versus classical separations result for "real-world" classes of probability distributions, via generically applicable quantum learning algorithms, which can run on near term devices. An ideal candidate for such a distribution class is precisely the output distributions of quantum circuits themselves - so called "quantum circuit Born machines". Given this, in the second part of the talk I will present a variety of results characterizing both the quantum and classical learnability (or non-learnability!) of the output distributions of quantum circuits, and discuss both the extent to which these results limit the potential advantages of near-term quantum generative modelling techniques, and the extent to which these results inform our understanding of the link between efficient simulation and efficient learning. Of particular interest are our results showing (a) the average-case hardness of learning sufficiently deep quantum circuit Born machines in the statistical query model, and (b) that while the output distributions of Clifford circuits can be efficiently learned, the addition of a single T gate (surprisingly!) renders the worst-case learning problem hard. |
Thursday, March 9, 2023 10:24AM - 11:00AM |
S69.00005: Understanding the role of data and quantum memory in a quantum learning landscape Invited Speaker: Jarrod McClean If we believe both that quantum computers may be able to do some computations exponentially faster than their classical counterparts and that we live in a quantum world, then our ability to learn from observational data as scientists may fundamentally change what we can do. Here, we will first review some recent results in quantum machine learning that allow us to put these ideas on a rigorous footing. We then show that quantum computers, and more specifically quantum memory, offer us an opportunity to learn from a quantum world with exponentially less data than traditional experiments. This exponential advantage holds in predicting properties of physical systems, performing quantum principal component analysis on noisy states, and learning approximate models of physical dynamics. Conducting experiments with up to 40 superconducting qubits and 1300 quantum gates, we demonstrate that a substantial quantum advantage can be realized using today's relatively noisy quantum processors. We then give an outlook on this technology and challenges that we face in expanding the reach of quantum technology in learning. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700