Current methods for classifying measurement trajectories in superconducting qubit systems produce fidelities systematically lower than those predicted by experimental parameters. Here, we place current classification methods within the framework of machine learning (ML) algorithms and improve on them by investigating more sophisticated ML approaches. We find that nonlinear algorithms and clustering methods produce significantly higher assignment fidelities that help close the gap to the fidelity possible under ideal noise conditions. Clustering methods group trajectories into natural subsets within the data, which allows for the diagnosis of systematic errors. We find large clusters in the data associated with T1 processes and show these are the main source of discrepancy between our experimental and ideal fidelities. These error diagnosis techniques help provide a path forward to improve qubit measurements.
Current methods for classifying measurement trajectories in superconducting qubit systems produce fidelities systematically lower than those predicted by experimental parameters. Here, we place current classification methods within the framework of machine learning (ML) algorithms and improve on them by investigating more sophisticated ML approaches. We find that nonlinear algorithms and clustering methods produce significantly higher assignment fidelities that help close the gap to the fidelity possible under ideal noise conditions. Clustering methods group trajectories into natural subsets within the data, which allows for the diagnosis of systematic errors. We find large clusters in the data associated with T1 processes and show these are the main source of discrepancy between our experimental and ideal fidelities. These error diagnosis techniques help provide a path forward to improve qubit measurements.
Accurate quantum state readout is crucial for error correction and algorithms, but measurement errors are detrimental. Readout fidelity is typically limited by a poor signal-to-noise ratio (SNR) and energy relaxation (T1 decay), a significant problem for superconducting qubits. While most approaches classify results using clustering algorithms on integrated readout signals, these methods cannot distinguish a qubit that was initially in the ground state from one that decayed to it during measurement. We instead propose using machine learning (ML) on the raw, non-integrated analog signal. We apply time-series classification models, such as a long short-term memory (LSTM) network, to the full data trajectory. We find that our LSTM model, combined with filtering and feature engineering, consistently outperforms clustering. The largest improvements come from reclassifying points in the boundary regions between clusters. These points correspond to atypical measurement records, likely due to transient or noisy features lost during data integration. By retaining temporal information, sequence-aware models like LSTMs can better discriminate these trajectories, whereas clustering methods based on integrated values are more prone to misclassification.
Abstract Measurement errors in quantum computers are very detrimental to quantum computations. The ability to efficiently and accurately readout quantum states is crucial for quantum error correction schemes and quantum algorithms. Readout fidelity is typically limited by a poor signal-to-noise (SNR) ratio between the quantum states we intend to classify, as well as energy relaxation (e.g., T1 decay) from an excited state to a lower state during readout. Superconducting quantum bits (qubit), one of the leading candidates for scalable quantum computing hardware, are particularly limited by energy relaxation due to their relatively short coherence times. While most approaches for classifying the results of readout on superconducting qubits typically utilize clustering algorithms (e.g., a Gaussian mixture model) on integrated readout signals, these cannot distinguish between a quantum bit that was in the ground state prior to measurement from a qubit that decays to the ground state during measurement. For this reason, we instead propose using machine learning (ML) on the raw (non-integrated) analog signal and classification models on the full time series data (i.e., the \emph{trajectory}). We observe that time series classification methods, such as our chosen long short-term memory (LSTM) model, in combination with filtering and feature engineering techniques, consistently outperform clustering models. In particular, we find that the largest improvements come from reclassifying points in the boundary regions between neighboring clusters. These boundary points correspond to measurement records that deviate from the typical cluster, likely due to transient or noisy features in the signal that are not captured when the data is integrated. By retaining temporal information, sequence-aware models such as LSTMs can better discriminate these trajectories, whereas clustering methods based on integrated values are more prone to misclassifications.
Accurate quantum state discrimination is critical for reliable qubit readout in current superconducting quantum processors, as it enables precise measurement outcomes, reduces readout-induced errors, and improves the overall reliability of quantum algorithm execution. Recently, machine learning (ML) techniques have been explored to enhance quantum state discrimination, outperforming traditional methods based on static thresholds. However, these approaches rely on hard-decision schemes that assign binary outcomes without accounting for readout uncertainty, thereby disregarding signal-level variations that can degrade measurement fidelity. This article introduces a measurement-aware classification approach based on Fuzzy C-means (FCM) clustering combined with data filtering, enabling soft-decision readout discrimination through confidence-based shot selection. Experimental validation was conducted on four different real IBM quantum devices, including single- and two-qubit readout scenarios, as well as multi-qubit circuits preparing GHZ and W-state with up to 10 qubits. Results show that the proposed method consistently outperforms state-of-the-art ML-based discriminators in terms of assignment fidelity and state reconstruction accuracy, demonstrating its effectiveness for enhancing readout quality in near-term quantum computing platforms. Quantitatively, the FCM-based discriminator yields 2–3% assignment fidelity improvements in single- and two-qubit readout and statistically significant gains in over 94% of GHZ and 97% of W multi-qubit experiments.
An important use-case for machine learning (ML) is that of determining readout results in quantum computers. In quantum computing (QC), classical ML models are currently being used to discriminate in-phase and quadrature (IQ) signal data to discriminate between quantum states, which is a fundamental QC operation. In our research we propose a Quantum K-Means (QK-Means) clustering technique to discriminate quantum states on the IBM Bogota quantum device, and compare its performance to the K-Means technique (its classical counterpart). We used both algorithms to perform a correlation analysis and probe cross-talk between couples of qubits on the device. We observed that QK-Means obtained test and training scores at par with the classical K-Means, and testing scores that were marginally lower than K-Means when the clusters weren’t visually separable. The training times for QK-Means were observed to be at par with an implementation of the K-Means that was not optimized. In this case, we concluded a weak correlation with a Pearson correlation coefficient of 0.2 on the (1, 2) and (2, 3) qubit couples. After analyzing the training scores, we also conclude that the 1 qubit has the worst performance at readout evidenced by the signal data not being visually separable and the low scores obtained on both clustering algorithms compared to the other qubits. Its poor performance is further verified by the calibration data showing a high readout error of 8,4%. This technique can be used to find correlations present in readout of quantum circuits and to determine cross-talk.
Abstract Advancements in the implementation of quantum hardware have enabled the acquisition of data that are intractable for emulation with classical computers. The integration of classical machine learning (ML) algorithms with these data holds potential for unveiling obscure patterns. Although this hybrid approach extends the class of efficiently solvable problems compared to using only classical computers, this approach has been realized for solving restricted problems because of the prevalence of noise in current quantum computers. Here, we extend the applicability of the hybrid approach to problems of interest in many-body physics, such as predicting the properties of the ground state of a given Hamiltonian and classifying quantum phases. By performing experiments with various error-reducing procedures on superconducting quantum hardware with 127 qubits, we managed to acquire refined data from the quantum computer. This enabled us to demonstrate the successful implementation of classical ML algorithms for systems with up to 44 qubits. Our results verify the scalability and effectiveness of the classical ML algorithms for processing quantum experimental data.
Quantum characterization, validation, and verification (QCVV) techniques are used to probe, characterize, diagnose, and detect errors in quantum information processors (QIPs). An important component of any QCVV protocol is a mapping from experimental data to an estimate of a property of a QIP. Machine learning (ML) algorithms can help automate the development of QCVV protocols, creating such maps by learning them from training data. We identify the critical components of "machine-learned" QCVV techniques, and present a rubric for developing them. To demonstrate this approach, we focus on the problem of determining whether noise affecting a single qubit is coherent or stochastic (incoherent) using the data sets originally proposed for gate set tomography. We leverage known ML algorithms to train a classifier distinguishing these two kinds of noise. The accuracy of the classifier depends on how well it can approximate the "natural" geometry of the training data. We find GST data sets generated by a noisy qubit can reliably be separated by linear surfaces, although feature engineering can be necessary. We also show the classifier learned by a support vector machine (SVM) is robust under finite-sample noise.