From Simulation to Experiment: Benchmarking Machine-Learning Models for Quantum-Dot State Recognition
Authors: Valeria Díaz Moreno, Ryan P Khalili, Daniel Schug, Patrick J. Walsh, Donovan L. Buterakos, Justyna P. Zwolak
Organizations: Department of Physics, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA · Department of Computer Science, University of Maryland, College Park, MD 20742, USA · Department of Applied Physics, Stanford University, Stanford, California 94305, USA · National Institute of Standards and Technology, Gaithersburg, MD 20899, USA · Joint Center for Quantum Information and Computer Science, University of Maryland, College Park, Maryland 20742, USA · Department of Physics, University of Maryland, College Park, Maryland 20742, USA
Machine-learning models for quantum-dot state recognition are commonly trained using synthetic data, where large labeled datasets can be generated efficiently. However, strong performance on simulated data does not guarantee equally strong performance in experiments. Here, we benchmark five neural network architectures for quantum-dot state recognition using a shared synthetic training dataset and independent synthetic and experimental test sets. We compare convolutional and fully connected neural networks, a mixture density network, a vision transformer, and a U-Net across multiple training set sizes and two input normalization schemes. We find that model performance and rankings are closely preserved when moving from held-out synthetic patches to previously unseen simulated devices, but change substantially when the same models are evaluated on experimental data. In particular, the architecture achieving the highest synthetic performance is not the one that performs best experimentally. Input normalization further affects training behavior and, for some architectures, synthetic-to-experimental transfer. These results show that synthetic benchmark performance alone is insufficient to select models for experimental deployment and highlight the need to jointly evaluate architecture, preprocessing, data efficiency, model complexity, and experimental transfer when developing machine-learning tools for automated quantum-dot characterization and control.
Figures & tables
Figure 1: Examples of (a) two large and (b) two small experimentally acquired charge stability diagrams included in the QFlow 2.0: Quantum dot data for machine learning dataset [ 14 ] . The manually assigned fractional state vector labels for the small CSDs are p(R1)=(0,0,0.1,0,0.9) and p(R2)=(0,0,1,0,0) .
Figure 2: (a) Example synthetic charge stability diagram and (b) the corresponding state map generated using QDFlow [ 18 , 21 ] . The state-map labels correspond to the five states possible for a double-QD device: no dot (ND), single left dot (SL), single central dot (SC), single right dot (SR), and double dot (DD). (c-i) Example patch R sampled from the CSD, highlighted in panel (a) with a white rectangle. (c-ii) Corresponding state-map patch, highlighted in panel (b) with a white rectangle. The fractional state vector assigned to this patch is p(R)=(0,0.1,0.5,0,0.4) .
Figure 3: Schematic representations of the five model architectures considered in this work. Each architecture receives the same 30\times30 preprocessed CSD patch as input and produces five output logits corresponding to the ND, SL, SC, SR, and DD states. The schematics illustrate the principal architectural differences among the models; implementation details are provided in section 2.4 .
Hyperparameter
MLP
CNN
MDN
ViT
U-Net
Learning rate η
1\times10−3
5\times10−4
1\times10−3
1\times10−4
5\times10−4
Weight decay
1\times10−4
2\times10−4
1\times10−4
3\times10−4
1\times10−4
Dropout
0.3
0
0.2
0.1
0.2
Activation
GELU
ReLU
GELU
GELU
ReLU
LR schedule
Cosine
Fixed
Cosine
Cosine
Cosine
Table 1: AdamW optimizer hyperparameters and learning-rate schedules used for the five model architectures.
Figure 4: MSE score, SMSE=1−MSE , as a function of synthetic training-data budget for models evaluated on (a) the held-out synthetic test set, (b) the device-held-out synthetic test set, and (c) the experimental test set. For panels (a) and (b), synthetic samples are importance-weighted such that the marginal distribution of dominant QD states matches that of the experimental test set; panel (c) shows the experimental data without reweighting. Within each architecture, points correspond to training-data budgets of 20% , 40% , 60% , 80% , and 100% . Colors identify the model architecture, while marker styles distinguish z-score (squares) and min-max (circles) normalization. Points show the mean performance across the 10 trained models for each condition, and error bars indicate one standard deviation across the 10 model-level scores, providing a measure of model-to-model variability.
Baseline
Held-out synthetic
Device-held-out synthetic
Experimental
Uniform
0.326
0.316
0.277
Training dataset mean
0.337
0.333
0.286
Table 2: MSE scores for the image-independent baseline predictors on the three evaluation datasets. The training dataset mean was established based on the full Synthetic test-set scores use the dominant-state importance weighting described in the text; the experimental test set is evaluated without reweighting.
Figure 5: (a) Number of training epochs before early stopping based on the validation KL-divergence loss for the MLP, CNN, MDN, ViT, and U-Net as a function of the synthetic training-data budget. Circles and squares denote min-max and z-score normalization, respectively; colors identify the model architectures and match those used in figure 3 . Points show the mean across the 10 trained models for each condition, and error bars indicate one standard deviation across the 10 model-level values, providing a measure of model-to-model variability in training behavior. (b) Number of trainable parameters for each architecture, reported in millions (M).
Figure 6: Predictive performance as a function of model complexity for (a) the held-out synthetic test set, (b) the device-held-out synthetic test set, and (c) the experimental test set. Mean MSE score, SMSE , is shown as a function of the number of trainable parameters for models trained using the full synthetic training-data budget. Circles and stars show the mean across the 10 trained models for each architecture using min-max and z-score normalization, respectively, with error bars indicating one standard deviation. Dashed and dotted lines connect the Pareto-optimal architectures for min-max and z-score normalization, respectively, when the corresponding frontier contains more than one model. The performance values used in the Pareto analysis follow the same evaluation procedure as in figure 4 : synthetic test-set scores are importance-weighted to match the experimental dominant-state distribution, while experimental scores are evaluated without reweighting.
Scaling semiconductor quantum dot arrays toward fault-tolerant quantum computing requires efficient tuneup of spin qubits, a process that depends on the analysis of charge stability maps (CSMs) and remains largely manual. While machine learning has been widely applied to CSM analysis in reservoir-coupled devices, automated tuning in the increasingly important isolated-mode regime has received limited attention. In isolated-mode CSMs, charge transitions appear as near-vertical lines, making them well suited to compact, task-specific models. We present two convolutional neural networks with fewer than one million parameters, trained on CSMs collected from 32 silicon metal-oxide-semiconductor (SiMOS) double-quantum-dot devices measured at approximately 1 K using an automated cryogenic probing system. Sixteen devices were used for training and sixteen were held out to evaluate cross-device generalization against hand-labeled ground truth. CSMClassifier identifies charge instability and sensor artifacts, achieving 94% macro-averaged accuracy across three quality classes on 2,407 held-out images. ChargeLineNet localizes charge-transition lines and determines electron occupancy, achieving 95.3% exact line-count accuracy on 1,131 held-out images. Combined into a single pipeline, the models correctly determine electron occupancy for 93.8% of clean held-out images. Pre-training on synthetic images substantially improves label efficiency. Fine-tuning the pre-trained model on limited experimental data maintains over 90% accuracy, whereas training from scratch degrades significantly under the same conditions. Together, the two models occupy only 6.5 MB and process images in less than 60 ms on standard laboratory hardware, demonstrating a practical path toward scalable, automated characterization and tuneup of quantum-dot devices.
Hyma Vallabhapurapu, Marco Candido, Krishna Choudhary +7
Diraq, Sydney, New South Wales, Australia · School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, New South Wales, Australia
The rapid growth of computer vision and increasingly complex image recognition tasks has exposed fundamental computational limitations of classical machine learning models, motivating the exploration of quantum computing as an emerging new paradigm. This paper presents a comprehensive benchmarking study of classical and quantum machine learning models for image recognition on the MNIST handwritten digit dataset, evaluating both traditional models, a Classical Support Vector Machine (CSVM) and a Quantum Support Vector Machine (QSVM), and deep neural network models, a Classical Convolutional Neural Network (CCNN) and a Quantum Convolutional Neural Network (QCNN), across four performance dimensions: classification accuracy, computational runtime, parameter count, and memory requirements. Experiments are conducted as functions of both feature dimensionality and sample size, and across CPU and GPU execution environments, providing a controlled, multidimensional comparison to address gaps in prior work. For the SVM-based models, QSVM consistently outperforms CSVM in accuracy, reaching ∼ 0.90 versus ∼ 0.85 at 1,000 samples, with a higher computational cost. A feature count of 10 qubits and a sample size in the range of 200 -- 500 emerge as practical operating points that balance accuracy and runtime. For the neural network models, CCNN and QCNN achieve comparable classification accuracy, both exceeding 0.96 at 64 features and 60,000 samples, yet QCNN offers substantially superior parameter and memory efficiency, requiring ∼ 94% fewer parameters and ∼ 75% less memory than CCNN at higher feature counts, while incurring higher runtime. Across both model families, quantum models consistently outperform classical models by greater margins in accuracy as feature dimensionality or sample size increases.
Sudip Vhaduri, Ryan Gammon, Sayanton Dibbo
Department of Computer Science, University of Alabama, AL 35487 · Clemson University
Efficiently characterizing quantum dot (QD) devices is a critical bottleneck when scaling quantum processors based on confined spins. Measuring high-resolution charge stability diagrams (or CSDs, data maps which crucially define the occupation of QDs) is time-consuming, particularly in emerging architectures where CSDs must be acquired with remote sensors that cannot probe the charge of the relevant dots directly. In this work, we present a generative approach to accelerate acquisition by reconstructing full CSDs from sparse measurements, using a conditional diffusion model. We evaluate our approach using two experimentally motivated masking strategies: uniform grid-based sampling, and line-cut sweeps. Our lightweight architecture, trained on approximately 9,000 examples, successfully reconstructs CSDs, maintaining key physically important features such as charge transition lines, from as little as 4% of the total measured data. We compare the approach to interpolation methods, which fail when the task involves reconstructing large unmeasured regions. Our results demonstrate that generative models can significantly reduce the characterization overhead for quantum devices, and provides a robust path towards an experimental implementation.
Vinicius Hernandes, Joseph Rogers, Rouven Koch +5
QuTech and Kavli Institute of Nanoscience, Delft University of Technology, Delft, the Netherlands