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 D73: Quantum Machine Learning IFocus
|
Hide Abstracts |
Sponsoring Units: DQI Chair: Prasanna Date, Oak Ridge National Lab Room: Room 405 |
Monday, March 6, 2023 3:00PM - 3:12PM |
D73.00001: Using Quantum Circuits with Convolutional Neural Networks for Multi-Object Detection and Classification Mandeep Saggi, Sabre Kais Multi-Object classification, detection, and recognition are the most often areas where it is already making an impact using the classical machine learning paradigm. Since the first quantum computers appeared and have allowed an exponential increase in the speed of solving NP-complete problems. There has been a rapid increase in interest in quantum algorithms for object recognition and detection. The goal of this paper is to offer a solution to the problem of research and development of quantum algorithms and methods on various real-time applications. The implementation of quantum-classical algorithms enables the conversion of a classical image into the quantum state, to predict the object localization through bounding box and multi-label classification. Quantum machine learning (QML) can process image data faster and more accurately than classical computers; potentially save costs in broader technology development, and support the current era of intermediate-scale quantum technology. |
Monday, March 6, 2023 3:12PM - 3:24PM |
D73.00002: Quantum natural language processing applications on high-performance computing systems and quantum devices Eduardo A Coello Perez, In-Saeng Suh, Prasanna Date, John P Gounley, Mayanka Chandra Shekar, Kathleen Hamilton Quantum natural language processing (QNLP) is a cutting-edge application aiming to develop NLP models to be executed on quantum computers. We assess the feasibility and accuracy of QNLP models using numerical simulators on HPC systems and actual quantum hardware. In particular, we use classical simulators and an hybrid HPC-quantum workflow to implement quantum pipelines and neural networks in combination with default datasets to demonstrate a QNLP application on HPC and NISQ devices. |
Monday, March 6, 2023 3:24PM - 3:36PM |
D73.00003: Parameterized quantum circuits for reinforcement learning of classical rare dynamics Sumeet Khatri, Alissa Wilms, Laura Ohff, Andrea Skolik, Jens Eisert In the study of non-equilibrium or industrial systems, rare events are crucial for understanding the systems' behavior and the effective search for such rare dynamics is frequently the subject of research. Since they are atypical, one requires specific methods for sampling and generating rare event statistics in an automated and statistically meaningful way. Recent publications have shown variational quantum algorithms to be among the most promising candidates for near-term applications on quantum devices. In this article, we propose two quantum reinforcement learning (QRL) approaches to study rare dynamics of time-dependent systems and investigate their benefits over classical approaches based on neural networks. We investigate how architectural choices such as different data encoding strategies and weights influence the successful learning by QRL agents and find a numerical separation in the benefits of data re-uploading for policy-based and value-based quantum approaches. We demonstrate that a quantum agent needs just a single qubit to be capable of learning and representing the rare dynamics of a random walker with a comparable performance of a simple neural network. Furthermore, we are able to numerically demonstrate an improved environment exploration during learning and a better performance in coping with environment scaling by the quantum agents in comparison to their classical counterparts. This is the first study of QRL in rare event statistics and suggests that QRL is a viable method to study rare dynamics of a system. |
Monday, March 6, 2023 3:36PM - 3:48PM |
D73.00004: Hybrid Classical-Quantum Machine Learning for Image Recognition on CIFAR-10 Julia Kwok, Nicholas S Shorter, Danielle M Couger, Joshua A Job, Steven H Adachi, Derek K Wise Even with the limitations of current noisy intermediate scale quantum (NISQ) devices, hybrid classical-quantum machine learning implementations have been demonstrated on both NISQ hardware and in simulation performing image classification. Building on previous work, input images' latent representations, coming from a classical neural network such as EfficientNet, are processed by a quantum circuit, whose measured outputs are then used by a classical network to classify input images. Improvements to prior hybrid methods are implemented and the resultant model trained and evaluated on the CIFAR-10 standard computer vision dataset. We present an overview of the theory behind these hybrid approaches, the improvements made to them, and a comparison of the results achieved from those improvements to top classical algorithms applied to the same data. |
Monday, March 6, 2023 3:48PM - 4:00PM |
D73.00005: Spectral asymmetry of quantum feature maps with real world classical data Mekena L Metcalf, Jorja J Kirk Expressing classical data on quantum Hamiltonians for near term quantum machine learning algorithms is expected to provide an advantage over classical algorithms, particularly for quantum kernel methods[1]. Recently, supervised and unsupervised learning demonstrated better accuracy performance using quantum kernels in classification tasks [2]. We evaluate the spectral composition of the data feature maps used to obtain the kernel coefficients with real world data. We find the spectral composition reveals characteristics between differing data types. In the case of binary classification, we find spectral asymmetry in the feature space which we can use to classify incoming data. Our results imply that real world classical data is expressed in different symmetry sectors of the Hilbert space. This results in a semi-supervised quantum learning task for binary classification where only one label is needed to estimate the label of incoming data which can be cast with quantum algorithm subroutines. We further find statistically significant ground state overlap for ZZ feature maps between data classes and establish performance metrics using overlap probabilities. |
Monday, March 6, 2023 4:00PM - 4:36PM |
D73.00006: The Complexity of NISQ Invited Speaker: Jordan Cotler The opportunities provided by contemporary quantum devices and the challenges they engender have invigorated theorists and experimentalists alike, ushering in a new age of research into quantum computation referred to as the NISQ (noisy intermediate-scale quantum) era. Computation in the NISQ era is facilitated by hybrid quantum-classical algorithms: a classical computer repeatedly runs a noisy quantum device with various gate sequences to obtain different classical output bitstrings, and then performs classical post-processing on those strings. This state of affairs suggests fundamental questions: How powerful are these NISQ algorithms compared to classical algorithms? Are NISQ algorithms inherently weaker than fault-tolerant quantum algorithms? In this talk we formalize and study these questions through the lens of computational complexity theory. Our findings suggest that NISQ algorithms can be super-polynomially more powerful than classical algorithms, but super-polynomially weaker than fault-tolerant quantum computation. We further analyze the power of NISQ algorithms for three well-studied problems: unstructured search, the Bernstein-Vazirani problem, and quantum state learning. Our proof techniques leverage and extend recent advances in quantum learning theory. |
Monday, March 6, 2023 4:36PM - 4:48PM |
D73.00007: Expressivity and generalization error of projected fidelity quantum kernels Beng Yee Gan, Supanut Thanasilp, Daniel Leykam, Dimitris G Angelakis Kernel methods solve nonlinear problems using linear models by mapping data into a higher dimensional feature space. While quantum computers can efficiently perform high dimensional feature maps, it has been shown that models based on global fidelity quantum kernels typically do not generalize well (i.e., perform poorly on unseen data) as the number of qubits increases. Projected quantum kernels have been proposed to resolve this generalization issue [1], but their properties remain largely unknown. Here, we study key properties of the projected quantum kernels including their expressivity and generalization error using kernel eigen-decomposition. We show that projected fidelity quantum kernels share a subset of their eigen-spectrum with global fidelity quantum kernels and how projected kernels impose bias on quantum models [2]. We also study how the acts of projection into smaller subspaces and composing these subspaces will affect the performance of quantum models. Finally, we analytically bound their generalization error by the difference between mean purities and mean embedding purities. Our work provides a deeper understanding of the properties of projected fidelity quantum kernels. |
Monday, March 6, 2023 4:48PM - 5:00PM |
D73.00008: Generalization of Quantum Kernel Methods Ruslan Shaydulin, Abdulkadir Canatar, Evan Peters, Cengiz Pehlevan, Stefan M Wild Quantum computers are known to provide speedups over classical state-of-the-art machine learning methods in some specialized settings. For example, quantum kernel methods have been shown to provide an exponential speedup on a learning version of the discrete logarithm problem. Understanding the generalization of quantum models is essential to realizing similar speedups on problems of practical interest. Recent results demonstrate that generalization is hindered by the exponential size of the quantum feature space. Although these results suggest that quantum models cannot generalize when the number of qubits is large, in this paper we show that these results rely on overly restrictive assumptions. We consider a wider class of models by varying a hyperparameter that we call quantum kernel bandwidth. We analyze the large-qubit limit and provide explicit formulas for the generalization of a quantum model that can be solved in closed form. Specifically, we show that changing the value of the bandwidth can take a model from provably not being able to generalize to any target function to good generalization for well-aligned targets. Our analysis shows how the bandwidth controls the spectrum of the kernel integral operator and thereby the inductive bias of the model. We demonstrate empirically that our theory correctly predicts how varying the bandwidth affects generalization of quantum models on challenging datasets, including those far outside our theoretical assumptions. We discuss the implications of our results for quantum advantage in machine learning. |
Monday, March 6, 2023 5:00PM - 5:12PM |
D73.00009: Numerical evidence against advantage with quantum fidelity kernels on classical data Lucas Slattery, Ruslan Shaydulin, Shouvanik Chakrabarti, Sami Khairy, Stefan M Wild, Marco Pistoia Quantum machine learning techniques are commonly considered one of the most promising candidates for demonstrating practical quantum advantage. In particular, quantum kernel methods have been demonstrated to be able to learn certain classically intractable functions efficiently if the kernel is well-aligned with the target function. In the more general case, quantum kernels are known to suffer from exponential ``flattening'' of the spectrum as the number of qubits grows, preventing generalization and necessitating the control of the inductive bias by hyperparameters. We show that the general-purpose hyperparameter tuning techniques proposed to improve the generalization of quantum kernels lead to the kernel becoming well-approximated by a classical kernel, removing the possibility of quantum advantage. We provide extensive numerical evidence for this phenomenon utilizing multiple previously studied quantum feature maps and both synthetic and real data. Our results show that unless novel techniques are developed to control the inductive bias of quantum kernels, they are unlikely to provide a quantum advantage on classical data. |
Monday, March 6, 2023 5:12PM - 5:24PM |
D73.00010: Exponential concentration and untrainability in quantum kernel methods Supanut Thanasilp, Samson Wang, Marco Cerezo, Zoe Holmes Kernel methods in Quantum Machine Learning (QML) have recently gained significant attention as a potential candidate for achieving a quantum advantage in data analysis. Among other attractive properties, when training a kernel-based model one is guaranteed to find the optimal model's parameters due to the convexity of the training landscape. However, this is based on the assumption that the quantum kernel can be efficiently obtained from a quantum hardware. In this work we study the trainability of quantum kernels from the perspective of the resources needed to accurately estimate kernel values. We show that, under certain conditions, values of quantum kernels over different input data can be exponentially concentrated (in the number of qubits) towards some fixed value, leading to an exponential scaling of the number of measurements required for successful training. We identify four sources that can lead to concentration including: the expressibility of data embedding, global measurements, entanglement and noise. For each source, an associated concentration bound of quantum kernels is analytically derived. Lastly, we show that when dealing with classical data, training a parametrized data embedding with a kernel alignment method is also susceptible to exponential concentration. Our results are verified through numerical simulations for several QML tasks. Altogether, we provide guidelines indicating that certain features should be avoided to ensure the efficient evaluation and the trainability of quantum kernel methods. |
Monday, March 6, 2023 5:24PM - 5:36PM |
D73.00011: Synergy Between Quantum Circuits and Tensor Networks: Short-cutting the Race to Practical Quantum Advantage (Part 1) Manuel S Rudolph, Jacob E Miller, Jing Chen, Atithi Acharya, Alejandro Perdomo-Ortiz Despite their success in "quantum supremacy" sampling tasks, the use of near-term quantum devices for solving high-value computational problems remains an open challenge. Proposals for achieving practical quantum advantage largely use parametrized quantum circuits (PQCs), but the existence of barren plateaus in the optimization landscape of these models makes training naively-initialized PQCs extremely difficult. We introduce a scalable means of leveraging classical resources to compute task-specific initializations for PQCs, which we show significantly boosts their performance on a variety of quantum simulation and machine learning problems, while demonstrably avoiding barren plateaus. Our method uses tensor network techniques to first identify a promising solution using available classical resources, before decomposing it into the parameters of a PQC. By utilizing a synergistic computing framework which blends quantum-inspired and fully-quantum models, our work opens new avenues for unlocking the full power of quantum devices for solving challenging real-world problems. |
Monday, March 6, 2023 5:36PM - 5:48PM |
D73.00012: Synergy Between Quantum Circuits and Tensor Networks: Short-cutting the Race to Practical Quantum Advantage (Part 2) Jacob E Miller, Manuel S Rudolph, Jing Chen, Atithi Acharya, Alejandro Perdomo-Ortiz Despite their success in "quantum supremacy" sampling tasks, the use of near-term quantum devices for solving high-value computational problems remains an open challenge. Proposals for achieving practical quantum advantage largely use parametrized quantum circuits (PQCs), but the existence of barren plateaus in the optimization landscape of these models makes training naively-initialized PQCs extremely difficult. We introduce a scalable means of leveraging classical resources to compute task-specific initializations for PQCs, which we show significantly boosts their performance on a variety of quantum simulation and machine learning problems, while demonstrably avoiding barren plateaus. Our method uses tensor network techniques to first identify a promising solution using available classical resources, before decomposing it into the parameters of a PQC. By utilizing a synergistic computing framework which blends quantum-inspired and fully-quantum models, our work opens new avenues for unlocking the full power of quantum devices for solving challenging real-world problems. |
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