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 B72: Holistic Benchmarking and Calibration of Quantum SystemsFocus
|
Hide Abstracts |
Sponsoring Units: DQI Chair: Megan Dahlhauser, Sandia National Laboratories Room: Room 406 |
Monday, March 6, 2023 11:30AM - 11:42AM |
B72.00001: Experimental demonstration of mirror circuit fidelity estimation Fernando A Calderon-Vargas, Stefan Seritan, Timothy J Proctor, Mohan Sarovar The verification of the accuracy with which a quantum computer can implement an algorithmic circuit is a crucial milestone towards reliable quantum computation. In arXiv:2204.07568, we introduced mirror circuit fidelity estimation (MCFE), a novel technique for quantifying the fidelity of quantum circuit executions on a specific quantum computer. In this work, we present the experimental demonstration of MCFE and compare its performance with similar techniques, like direct fidelity estimation. |
Monday, March 6, 2023 11:42AM - 11:54AM |
B72.00002: Estimating the Error Budgets of Quantum Gates in Superconducting Qubits Miha Papic, Inés de Vega There is a variety of different error mechanisms present in current superconducting based quantum computers and distinguishing between their contributions can prove to be challenging. However, an accurate estimate of the main error source contributions is a valuable resource which enables the efficient selection of the appropriate calibration or mitigation procedures. |
Monday, March 6, 2023 11:54AM - 12:06PM |
B72.00003: K-body Pauli error estimation Robin Harper Learning the Pauli error rates of arbitrary channels remains an important aspirational aim for multi-qubit devices. Such information is vital in error mitigation as well as the design of codes and bespoke decoders for error correction, especially as quantum devices inch towards full error correcting thresholds. Clearly it is not possible to learn (or even represent) the full Pauli error channel for devices much larger than, say, 20 qubits. In order to maintain efficiency, existing protocols require some assumptions, typically assuming the Pauli channel is sparse or k-local. Recently Flammia and O'Donnell showed how to recover all Pauli error terms of an arbitrary Pauli error channel to a precision of epsilon using population recovery techniques. Here we build on that work, seeking to retain a practical implementation but removing measurement error and increasing relative precision; albeit at the cost of limiting the recovery to Pauli error terms that act non-trivially on only k or fewer qubits. |
Monday, March 6, 2023 12:06PM - 12:18PM |
B72.00004: A physics-based machine learning approach to quantum device characterization Daniel Hothem, Kevin Young, Timothy J Proctor, Tommie Catanach Modern quantum computers are noisy and error-prone. They are also growing rapidly, necessitating the development of scalable methods for predicting their performance. Although unobservable, estimations of process fidelity provide one path to scalable benchmarks. In this work, we use existing scalable methods for estimating process fidelity to train neural networks to predict the process fidelity of unknown circuits of interest. As off-the-shelf neural networks are agnostic to the underlying physics present in a quantum device, we augment their training with information from traditional physics-based modelling approaches, ultimately allowing these neural networks to serve as device proxies for quick and accurate querying and estimation of a device’s performance. |
Monday, March 6, 2023 12:18PM - 12:30PM |
B72.00005: Reinforcement learning for calibrating quantum processors Kevin Young As quantum processors grow larger, manual calibration of logic operations quickly becomes impossible. Instead, automated solutions are required that can rapidly identify miscalibrated control parameters and correct them. Reinforcement learning has demonstrated remarkable success at automating a wide range of tasks through trial and error and feedback, providing nearly turn-key methods that efficiently balance parameter space exploration with exploitation of gained knowledge. In this talk, I will outline a simple example of how calibration of a quantum device can be cast as an environment/agent/reward problem that yields to techniques from deep reinforcement learning. Our approach probes the calibration state of the system using long, periodic gate sequences that are maximally sensitive to coherent errors---those that are most susceptible to calibration. I demonstrate the performance of this technique in simulation. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525. |
Monday, March 6, 2023 12:30PM - 12:42PM |
B72.00006: Error mitigation via adaptive calibrations Yonatan Cohen, Lior Ella Feedback and adaptivity are playing an increasingly important role in quantum circuit execution[1-3]. These mechanisms can be applied both during circuit execution, where timing requirements are extremely stringent due to limited coherence times, as well as between circuit executions, where timing requirements play an important role due to 1/f noise and parameter drift. In this talk, we show how algorithms such as iterative phase estimation, entangled 2-qubit states and iterative state preparation techniques can be used as highly-sensitive probes to significantly speed up calibration and characterization workflows for superconducting qubit platforms. This results in higher circuit performance due to the ability to track drifting qubit parameters with much less overhead than traditional, non-feedback based methods. Furthermore, these ideas can be expanded to be used during quantum error correction cycles, to detect and prevent in-situ parameter drift during ancilla measurement rounds. |
Monday, March 6, 2023 12:42PM - 12:54PM |
B72.00007: Characterizing Continuously Parameterized Gates with Gate Set Tomography Jordan Hines, Corey I Ostrove, Stefan Seritan, Erik Nielsen, Kevin Young, Robin J Blume-Kohout, Timothy J Proctor Gate set tomography (GST) is a widely used technique for characterizing a set of noisy quantum gates. GST is formulated for a discrete gate set, but quantum processors often use gates with continuous-valued parameters (such as rotation angles), and the error on these gates may depend on the values of their parameters. Here, we present a method for GST of gate sets containing continuously parameterized gates. We introduce a class of parameter-dependent models for error on continuously parameterized single-qubit gates, using the error generator formalism, and we show how to use GST to fit these models to data. The result is an estimated error map, for each gate, that is a function of the gate's parameters. We demonstrate our method with single-qubit GST experiments, and we explore how well our error models capture real device noise. |
Monday, March 6, 2023 12:54PM - 1:06PM |
B72.00008: Efficient experimental verification of continuously-parameterized quantum gates via randomized analog verification Ryan Shaffer, Hang Ren, Emiliia Dyrenkova, Christopher G Yale, Daniel S Lobser, Ashlyn D Burch, Matthew N Chow, Melissa C Revelle, Susan M Clark, Hartmut Haeffner Near-term quantum computers implement circuits directly using the physical native gate set of the device. These gates often have a parameterization (e.g., rotation angles) which enables a continuous range of operations. Verification of the correct operation of these gates across the allowable range of parameters is important for gaining confidence in the reliability of these devices. In this work, we demonstrate the application of the randomized analog verification (RAV) procedure for efficient verification of continuously-parameterized quantum gates. This procedure involves generating random sequences of randomly-parameterized layers of gates chosen from the native gate set of the device, and then stochastically compiling an approximate inverse to this sequence such that executing the full sequence on the device should leave the system near its initial state. We show that fidelity estimates made via RAV have a lower variance than fidelity estimates made via cross-entropy benchmarking (XEB), which thus provides an efficiency advantage when estimating the error rate to some desired precision. |
Monday, March 6, 2023 1:06PM - 1:18PM |
B72.00009: Optimized Bayesian System Identification in Quantum Devices Ashish Kakkar, Thomas M Stace, Pranav S Mundada, Yuval Baum, Jiayin Chen, Li Li, Victor Perunicic, Michael Hush, Ting Rei Tan, Christophe Valahu, Michael Biercuk Identifying and calibrating quantitative dynamical models for physical quantum systems is critical in the development of a quantum computer. We present a closed-loop Bayesian learning algorithm for estimating multiple unknown parameters in a dynamical model, using optimised experimental “probe” controls and measurement. The estimation algorithm is based on a Bayesian particle filter, and is designed to autonomously choose informationally-optimised probe experiments with which to compare to model predictions. In both simulated and experimental demonstrations, we see that successively longer pulses are selected as the posterior uncertainty iteratively decreases, leading to an exponential scaling in the accuracy of model parameters with the number of experimental queries. In an experimental calibration of a single qubit ion trap, we achieve parameter estimates in agreement with standard calibration approaches but requiring ∼ 20× fewer experimental measurements. We also demonstrate the performance of the algorithm on multi qubit superconducting devices, demonstrating the flexibility of these techniques. |
Monday, March 6, 2023 1:18PM - 1:54PM |
B72.00010: Measuring progress towards useful quantum computation using precision benchmarks Invited Speaker: Timothy J Proctor Measuring a quantum computer’s capability, i.e., learning what circuits it can run with high fidelity, is essential for understanding that device’s computational power and for tracking technological progress. However, now that state-of-the-art quantum computers can run circuits that are infeasible to classically simulate, even directly verifying that a quantum computer has executed a specific computation with some given fidelity has become practically impossible (for general computations). In this talk, I will present methods for learning a quantum computer’s capability that are efficient and scalable. I will show how a method called mirror circuit fidelity estimation can be used to efficiently estimate the fidelity with which a quantum computer can implement any given circuit, on any number of qubits. I will then show how this method can be used to create efficient and principled algorithmic benchmarks that measure progress towards implementing a full-scale algorithm. To fully understand a quantum computer’s capability, we must go beyond measuring a processor’s execution fidelity for one or more given circuits: we need to predict this fidelity for general circuits. I will discuss methods for constructing models for a quantum computer’s capability, including conventional error models as well as machine learned surrogates, and highlight the main scientific challenges in creating accurate capability models. |
Monday, March 6, 2023 1:54PM - 2:06PM |
B72.00011: Autonomous calibration of quantum processors with Bayesian filtering John P Marceaux, Kevin Young Autonomous calibration of emerging many qubit quantum processors will be essential to achieving and maintaining sufficiently small error rates to enable fault tolerant quantum computation. Classic nonlinear Kalman filters provide one possible framework, and we have found that quantum gate set tomography is amenable to estimation with the extended and sigma point filters, achieving fits comparable with maximum likelihood estimation in simulation. This compatibility indicates a promising route towards online and adaptive calibration of quantum processors using existing techniques in classical control theory. We incorporate the effect of controls in a filter’s dynamic state model to derive an optimal linear quadratic regulator for a quantum gate set. After selecting a noise model and a control map for a given hardware, our regulator may be deployed and called periodically for gate tuneup or continuously for drift compensation. Improvements in metrics like average gate set infidelity or randomized benchmarking rates verify successful calibration cycles. We are constructing a low level, hardware agnostic calibration algorithm for the next generation of quantum computers.
|
Monday, March 6, 2023 2:06PM - 2:18PM |
B72.00012: Designing quantum gates using deep reinforcement learning Ho Nam Nguyen, Marin Bukov, Markus Schmitt, Felix Motzoi, Mekena Metcalf The advantage of a quantum computer over its classical counterpart relies heavily on the ability to perform high-fidelity quantum logic operations. Theoretical and empirical studies of error sources have resulted in many promising designs for the standard single-qubit and two-qubit gates. While these candidates capture the majority of the dynamics, unknown error processes in a realistic hardware are not explicitly addressed but implicitly via frequent calibration. In this work, we task a deep reinforcement learning agent to interact with a simulated quantum environment of superconducting transmon qubits to directly design quantum gates suitable to the true dynamics. With a learning objective based on the worst-case fidelity, instead of the commonly used average fidelity, our agent explores the vast design landscape of piecewise-constant pulses and finds non-trivial solutions for single-qubit rotation and cross-resonance entangling operation. |
Monday, March 6, 2023 2:18PM - 2:30PM |
B72.00013: Automated Model Selection for Gate Set Tomography Stefan Seritan, Kenneth Rudinger, Timothy J Proctor, Robin Blume-Kohout Gate set tomography (GST) is a characterization technique that provides high accuracy estimates of a quantum processor’s operations. Traditional GST models each gate as a dense process matrix, which can be prohibitively expensive to fit for multiple qubits and difficult to interpret. Reduced models that only contain a subset of the model parameters can mitigate both issues, but model selection, i.e. choosing which parameters to keep, remains an open problem. To that end, we present algorithms for performing automated model selection, allowing the construction of reduced models without needing a priori knowledge of device physics. We demonstrate our method with both simulations and by comparing against hand-tuned reduced models from existing experiments. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525. |
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