Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2604.24912

Data-Driven Hamiltonian Reduction for Superconducting Qubits via Meta-Learning

Arielle Sanford, Andrew T. Kamen, Frederic T. Chong, Andy J. Goldschmidt

quant-phcs.LGSubmitted Apr 27, 2026• Announced Apr 29, 2026• 0 citations
arXivHTMLPDF

Abstract

We introduce HAML (Hamiltonian Adaptation via Meta-Learning), a framework for fast online adaptation of effective Hamiltonian models of superconducting quantum processors. HAML proceeds in two phases. A supervised training phase uses an ensemble of simulated devices to learn an offline map from control inputs and device parameters to effective Hamiltonian coefficients. An online adaptation phase then uses a small number of hardware-accessible measurements to identify the unknown parameters of a new device. By training directly against effective two-qubit coefficients extracted from full multi-mode simulations, HAML implicitly learns the reduction from full multi-mode Hamiltonians to effective qubit descriptions without invoking perturbation theory. We further show that a variance-maximizing greedy selection of measurement configurations boosts online adaptation efficiency. We demonstrate HAML on a transmon-coupler-transmon system, recovering effective two-qubit coefficients across a wide range of operating regimes, including parameter regions where Schrieffer-Wolff perturbation theory (SWPT) breaks down. This establishes a scalable, sample-efficient approach to Hamiltonian reduction and characterization for near-term quantum processors, with direct implications for calibration, control, and error mitigation.

Related papers

Data-efficient quantum noise modeling via machine learning

Yanjun Ji, Marco Roth, David A. Kreplin +2 more

Maximizing the computational utility of near-term quantum processors requires predictive noise models that inform robust, noise-aware compilation and error mitigation. Conventional models often fail to capture the complex error dynamics of real hardware or require prohibitive characterization overhead. We introduce a data-efficient framework that first constructs a physically motivated, parameterized noise model, and subsequently employs machine-learning-driven Bayesian optimization to identify its parameters. Our approach circumvents costly characterization protocols by estimating algorithm- and hardware-specific error parameters directly from readily available experimental data derived from existing application and benchmark circuit executions. The generality and robustness of the framework are demonstrated across diverse algorithms and superconducting devices, yielding high-fidelity predictions by estimating an independent parameter set tailored to each specific algorithm-hardware context. Crucially, we show that a model calibrated exclusively on small-scale circuits accurately predicts the behavior of larger validation circuits. Our data-efficient approach achieves up to a 65% improvement in model fidelity quantified by the Hellinger distance between predicted and experimental circuit output distributions, compared to standard noise models derived from device properties. This work establishes a practical paradigm for application-aware noise characterization, enabling compilation and error-mitigation strategies tailored to the specific interplay between quantum algorithms and device-specific noise dynamics.

arXiv:2509.12933•Physical Review Applied•Feb 11, 2026•5 citations

Accelerated characterization of two-level systems in superconducting qubits via machine learning

Rama Khalil

Two-level systems (TLS) in superconducting quantum devices are a dominant source of decoherence, noise, and device variability, ultimately limiting qubit performance and scalability. Efficient and reliable extraction of TLS parameters from spectroscopy measurements is therefore essential for both device characterization and materials optimization. In this thesis, machine learning methods are developed to accelerate and enhance TLS characterization from superconducting qubit spectroscopy data. The approach combines physics-based simulation of driven qubit–TLS dynamics with supervised learning models trained on synthetic datasets to infer underlying physical parameters from two-dimensional spectroscopy maps. Particular emphasis is placed on understanding how different parameters are encoded in observable features, and on evaluating the extent to which the inverse problem is well-posed. The results demonstrate that machine-learning-assisted inference can significantly reduce the computational cost of parameter estimation compared to conventional fitting-based approaches, while maintaining high accuracy for spectrally encoded parameters such as TLS frequency and coupling strength. In contrast, decoherence parameters are shown to be intrinsically more challenging to extract from short-time spectroscopy data, highlighting fundamental limitations of the measurement protocol rather than the learning model. Overall, this work establishes a data-driven framework for scalable TLS characterization and provides insight into the interplay between physical observability and machine learning-based inference in quantum systems.

•Jan 1, 2026•0 citations

Physics-Informed Artificial Intelligence for Inverse Design and Decoherence Mitigation in Superconducting Transmon Architectures

Ravi Raja Ramisetty, gopi chand kudipudi, Anvith Aravapalli

Scaling superconducting quantum processors is fundamentally limited by environmentally induced decoherence arising from nanoscale fabrication defects. Conventional heuristic microwave control protocols cannot dynamically adapt to device-specific material variations, while purely data-driven deep learning approaches often violate physical constraints. Here we introduce a compact quantum physics-informed neural network (QPINN) that learns device-aware microwave control pulses for superconducting qubits. Trained on empirical defect characteristics extracted from 162 InAs nanosheet Josephson junction measurements, the model embeds the Lindblad master equation directly into its loss function to enforce physical consistency. Open-system simulations demonstrate a 42% reduction in leakage to non-computational states. Hardware experiments executed on the 156-qubit IBM Quantum processor show an increase in Ramsey dephasing time from 83.3 ± 7.8 μs to 113.0 ± 1.1 μs while maintaining gate fidelity exceeding 99.99%.

•Mar 15, 2026•0 citations

Active learning of quantum system Hamiltonians yields query advantage

Arkopal Dutt, Edwin Pednault, Chai Wah Wu +4 more

Hamiltonian learning is an important procedure in quantum system identification, calibration, and successful operation of quantum computers. Through queries to the quantum system, this procedure seeks to obtain the parameters of a given Hamiltonian model and description of noise sources. Standard techniques for Hamiltonian learning require careful design of queries and $O({\ensuremath{\epsilon}}^{\ensuremath{-}2})$ queries in achieving learning error $\ensuremath{\epsilon}$ due to the standard quantum limit. With the goal of efficiently and accurately estimating the Hamiltonian parameters within learning error $\ensuremath{\epsilon}$ through minimal queries, we introduce an active learner that is given an initial set of training examples and the ability to interactively query the quantum system to generate new training data. To ensure applicability on near-term quantum hardware, the active learner operates in batch mode as opposed to sequentially, proposing batches of queries to be made during learning. We formally specify and experimentally assess the performance of this Hamiltonian active learning (HAL) algorithm for learning the six parameters of a two-qubit cross-resonance Hamiltonian on four different superconducting IBM quantum devices. Compared with standard techniques for the same problem and a specified learning error, HAL achieves more than a $95%$ reduction in queries required, and upwards of $33%$ reduction over a sequential active learner. Moreover, with access to prior information on a subset of Hamiltonian parameters and given the ability to select queries with linearly (or exponentially) longer system interaction times during learning, HAL can exceed the standard quantum limit and achieve Heisenberg (or super-Heisenberg) limited convergence rates during learning.

arXiv:2112.14553•Physical Review Research•Jul 28, 2023•23 citations

Learning Lindblad Dynamics of a Superconducting Quantum Processor

Johann Bock Severin, Malthe A. Marciniak, Rune Thinggaard Birke +13 more

Accurate models of quantum processors are essential for understanding, calibrating, and improving their performance. In practice, model construction must balance physical detail against the experimental and computational effort required to reliably learn parameters. Compact descriptions therefore often rely on assumptions about which interactions, noise processes, or hidden degrees of freedom are relevant. Here we introduce LIMINAL, a data-driven framework for testing such assumptions and selecting minimal adequate Lindblad models. LIMINAL fits nested candidate models to time-resolved tomographic data and uses likelihood-ratio tests to decide when added physical mechanisms are warranted. We apply LIMINAL to a five-qubit superconducting processor, identifying an idling model with three-local Hamiltonian terms and two-local dissipation, while finding no support for three-local dissipation. We further apply it to recover driven single-qubit Hamiltonians, reconstruct a shaped-pulse Hamiltonian without assuming an analytic pulse model, and test hidden-qubit extensions in coupler-mediated dynamics, demonstrating the applicability of the framework for a wide range of tasks.

quant-pharXiv:2605.00626•May 4, 2026•3 citations

Robustly learning the Hamiltonian dynamics of a superconducting quantum processor

Dominik Hangleiter, Ingo Roth, Jonáš Fuksa +2 more

Abstract Precise means of characterizing analog quantum simulators are key to developing quantum simulators capable of beyond-classical computations. Here, we precisely estimate the free Hamiltonian parameters of a superconducting-qubit analog quantum simulator from measured time-series data on up to 14 qubits. To achieve this, we develop a scalable Hamiltonian learning algorithm that is robust against state-preparation and measurement (SPAM) errors and yields tomographic information about those SPAM errors. The key subroutines are a novel super-resolution technique for frequency extraction from matrix time-series, tensorESPRIT, and constrained manifold optimization. Our learning results verify the Hamiltonian dynamics on a Sycamore processor up to sub-MHz accuracy, and allow us to construct a spatial implementation error map for a grid of 27 qubits. Our results constitute an accurate implementation of a dynamical quantum simulation that is precisely characterized using a new diagnostic toolkit for understanding, calibrating, and improving analog quantum processors.

arXiv:2108.08319•Nature Communications•Nov 6, 2024•56 citations
We introduce HAML (Hamiltonian Adaptation via Meta-Learning), a framework for fast online adaptation of effective Hamiltonian models of superconducting quantum processors. HAML proceeds in two phases. A supervised training phase uses an ensemble of simulated devices to learn an offline map from control inputs and device parameters to effective Hamiltonian coefficients. An online adaptation phase then uses a small number of hardware-accessible measurements to identify the unknown parameters of a new device. By training directly against effective two-qubit coefficients extracted from full multi-mode simulations, HAML implicitly learns the reduction from full multi-mode Hamiltonians to effective qubit descriptions without invoking perturbation theory. We further show that a variance-maximizing greedy selection of measurement configurations boosts online adaptation efficiency. We demonstrate HAML on a transmon-coupler-transmon system, recovering effective two-qubit coefficients across a wide range of operating regimes, including parameter regions where Schrieffer-Wolff perturbation theory (SWPT) breaks down. This establishes a scalable, sample-efficient approach to Hamiltonian reduction and characterization for near-term quantum processors, with direct implications for calibration, control, and error mitigation.
Maximizing the computational utility of near-term quantum processors requires predictive noise models that inform robust, noise-aware compilation and error mitigation. Conventional models often fail to capture the complex error dynamics of real hardware or require prohibitive characterization overhead. We introduce a data-efficient framework that first constructs a physically motivated, parameterized noise model, and subsequently employs machine-learning-driven Bayesian optimization to identify its parameters. Our approach circumvents costly characterization protocols by estimating algorithm- and hardware-specific error parameters directly from readily available experimental data derived from existing application and benchmark circuit executions. The generality and robustness of the framework are demonstrated across diverse algorithms and superconducting devices, yielding high-fidelity predictions by estimating an independent parameter set tailored to each specific algorithm-hardware context. Crucially, we show that a model calibrated exclusively on small-scale circuits accurately predicts the behavior of larger validation circuits. Our data-efficient approach achieves up to a 65% improvement in model fidelity quantified by the Hellinger distance between predicted and experimental circuit output distributions, compared to standard noise models derived from device properties. This work establishes a practical paradigm for application-aware noise characterization, enabling compilation and error-mitigation strategies tailored to the specific interplay between quantum algorithms and device-specific noise dynamics.
Two-level systems (TLS) in superconducting quantum devices are a dominant source of decoherence, noise, and device variability, ultimately limiting qubit performance and scalability. Efficient and reliable extraction of TLS parameters from spectroscopy measurements is therefore essential for both device characterization and materials optimization. In this thesis, machine learning methods are developed to accelerate and enhance TLS characterization from superconducting qubit spectroscopy data. The approach combines physics-based simulation of driven qubit–TLS dynamics with supervised learning models trained on synthetic datasets to infer underlying physical parameters from two-dimensional spectroscopy maps. Particular emphasis is placed on understanding how different parameters are encoded in observable features, and on evaluating the extent to which the inverse problem is well-posed. The results demonstrate that machine-learning-assisted inference can significantly reduce the computational cost of parameter estimation compared to conventional fitting-based approaches, while maintaining high accuracy for spectrally encoded parameters such as TLS frequency and coupling strength. In contrast, decoherence parameters are shown to be intrinsically more challenging to extract from short-time spectroscopy data, highlighting fundamental limitations of the measurement protocol rather than the learning model. Overall, this work establishes a data-driven framework for scalable TLS characterization and provides insight into the interplay between physical observability and machine learning-based inference in quantum systems.
Scaling superconducting quantum processors is fundamentally limited by environmentally induced decoherence arising from nanoscale fabrication defects. Conventional heuristic microwave control protocols cannot dynamically adapt to device-specific material variations, while purely data-driven deep learning approaches often violate physical constraints. Here we introduce a compact quantum physics-informed neural network (QPINN) that learns device-aware microwave control pulses for superconducting qubits. Trained on empirical defect characteristics extracted from 162 InAs nanosheet Josephson junction measurements, the model embeds the Lindblad master equation directly into its loss function to enforce physical consistency. Open-system simulations demonstrate a 42% reduction in leakage to non-computational states. Hardware experiments executed on the 156-qubit IBM Quantum processor show an increase in Ramsey dephasing time from 83.3 ± 7.8 μs to 113.0 ± 1.1 μs while maintaining gate fidelity exceeding 99.99%.
Hamiltonian learning is an important procedure in quantum system identification, calibration, and successful operation of quantum computers. Through queries to the quantum system, this procedure seeks to obtain the parameters of a given Hamiltonian model and description of noise sources. Standard techniques for Hamiltonian learning require careful design of queries and $O({\ensuremath{\epsilon}}^{\ensuremath{-}2})$ queries in achieving learning error $\ensuremath{\epsilon}$ due to the standard quantum limit. With the goal of efficiently and accurately estimating the Hamiltonian parameters within learning error $\ensuremath{\epsilon}$ through minimal queries, we introduce an active learner that is given an initial set of training examples and the ability to interactively query the quantum system to generate new training data. To ensure applicability on near-term quantum hardware, the active learner operates in batch mode as opposed to sequentially, proposing batches of queries to be made during learning. We formally specify and experimentally assess the performance of this Hamiltonian active learning (HAL) algorithm for learning the six parameters of a two-qubit cross-resonance Hamiltonian on four different superconducting IBM quantum devices. Compared with standard techniques for the same problem and a specified learning error, HAL achieves more than a 9595%95 reduction in queries required, and upwards of 3333%33 reduction over a sequential active learner. Moreover, with access to prior information on a subset of Hamiltonian parameters and given the ability to select queries with linearly (or exponentially) longer system interaction times during learning, HAL can exceed the standard quantum limit and achieve Heisenberg (or super-Heisenberg) limited convergence rates during learning.
Accurate models of quantum processors are essential for understanding, calibrating, and improving their performance. In practice, model construction must balance physical detail against the experimental and computational effort required to reliably learn parameters. Compact descriptions therefore often rely on assumptions about which interactions, noise processes, or hidden degrees of freedom are relevant. Here we introduce LIMINAL, a data-driven framework for testing such assumptions and selecting minimal adequate Lindblad models. LIMINAL fits nested candidate models to time-resolved tomographic data and uses likelihood-ratio tests to decide when added physical mechanisms are warranted. We apply LIMINAL to a five-qubit superconducting processor, identifying an idling model with three-local Hamiltonian terms and two-local dissipation, while finding no support for three-local dissipation. We further apply it to recover driven single-qubit Hamiltonians, reconstruct a shaped-pulse Hamiltonian without assuming an analytic pulse model, and test hidden-qubit extensions in coupler-mediated dynamics, demonstrating the applicability of the framework for a wide range of tasks.
Abstract Precise means of characterizing analog quantum simulators are key to developing quantum simulators capable of beyond-classical computations. Here, we precisely estimate the free Hamiltonian parameters of a superconducting-qubit analog quantum simulator from measured time-series data on up to 14 qubits. To achieve this, we develop a scalable Hamiltonian learning algorithm that is robust against state-preparation and measurement (SPAM) errors and yields tomographic information about those SPAM errors. The key subroutines are a novel super-resolution technique for frequency extraction from matrix time-series, tensorESPRIT, and constrained manifold optimization. Our learning results verify the Hamiltonian dynamics on a Sycamore processor up to sub-MHz accuracy, and allow us to construct a spatial implementation error map for a grid of 27 qubits. Our results constitute an accurate implementation of a dynamical quantum simulation that is precisely characterized using a new diagnostic toolkit for understanding, calibrating, and improving analog quantum processors.