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

International Journal of Mathematic Exploration and Computer Education

When Is Machine Learning Needed for Quantum Memory Budget Prediction? A Baseline Study with Extreme-Scenario Boundary Exploration

David Vesterlund

Published Oct 19, 2026• 0 citations
Publisher

Abstract

Findings: On the original quantum-network dataset, the best nonlinear model, Random Forest, reduces mean absolute error (MAE) by only 4.0% relative to Ridge regression with one-hot-encoded categorical features (2.500 versus 2.605), while Ridge reduces MAE by 49.5% relative to a constant predictor (5.160). Nearly all predictive signal is associated with memory family; every individual numerical feature has |r| < 0.08 with the target memory budget mc. In four of five controlled synthetic scenarios—threshold effects, saturation, phase transitions, and family-specific nonlinearity—the best tree ensemble reduces MAE by 31–66% relative to Ridge. In the heterogeneous scenario, Random Forest also reduces the observed under-dimensioning frequency from 55.5% to 35.0%. For a two-way numerical interaction, however, the improvement is only 12.5%. These single-split results support Ridge with one-hot encoding as a necessary baseline, but do not establish a universal model-selection threshold. Method: The original dataset contains 1,000 seed-averaged entanglement-buffering configurations spanning six memory families, nine topology classes, three noise levels, seven traffic levels, and five network sizes (N ϵ {10,15,20,30,50}), with mc ϵ [1,20]. We compare a constant predictor, single-feature linear regression, multivariable linear regression, Ridge regression, Random Forest, and XGBoost on one fixed 60/20/20 train/validation/test split. MAE is the primary metric, supplemented by tolerance accuracy and the fraction of predictions below the target. To probe where nonlinear models begin to add value, we construct five synthetic datasets of 1,000 observations each with controlled threshold, interaction, saturation, phase-transition, and family-heterogeneous response functions. Because all reported comparisons use one split and seed, differences of a few percentage points are exploratory.

DOI
10.47485/3069-9703.1029

Related papers

Evaluating Quantum Support Vector Regression Methods for Price Forecasting Applications

Horst Stühler, Daniel Pranjić, Christian Tutschku

Support vector machines are powerful and frequently used machine learning methods for classification and regression tasks, which rely on the construction of kernel matrices. While crucial for the performance of this machine learning approach, choosing the most suitable kernel is highly problem-dependent. The emergence of quantum computers and quantum machine learning techniques provides new possibilities for generating powerful quantum kernels. Within this work, we solve a real-world price forecasting problem using fidelity and projected quantum kernels, which are promising candidates for the utility of near-term quantum computing. In our analysis, we examine and validate the most auspicious quantum kernels from literature and compare their performance with an optimized classical kernel. Unlike previous work on quantum support vector machines, our dataset includes categorical features that need to be encoded as numerical features, which we realize by using the one-hot-encoding scheme. One-hot-encoding, however, increases the dimensionality of the dataset significantly, which collides with the current limitations of noisy intermediate scale quantum computers. To overcome these limitations, we use autoencoders to learn a low-dimensional representation of the feature space that still maintains the most important information of the original data. To examine the impact of autoencoding, we compare the results of the encoded date with the results of the original, unencoded dataset. We could demonstrate that quantum kernels are comparable to or even better than the classical support vector machine kernels regarding the mean absolute percentage error scores for both encoded and unencoded datasets.

International Conference on Agents and Artificial Intelligence (ICAART)•Jan 1, 2024•7 citations

Memory Physics, Finite Capacity, and Topology in Heterogeneous Quantum Networks

David Vesterlund

Abstract Quantum-network capacity planning couples memory physics, finite storage, topology, and traffic. We study these interactions using a discrete-event simulator with six representative memory profiles. A memory–topology atlas contains 9,360 completed runs, complemented by 8,730 confirmatory runs and an extended per-node memory-budget sweep. We distinguish a threshold relative to the best performance observed on a finite grid from a physical saturation limit. Among 840 reported threshold estimates, 22.3% reach 95% of the observed maximum by four slots per node, while 35.5% first reach it at the upper grid boundary of 64 slots. Four tested scalar reductions do not provide a topology independent collapse; the best pooled quadratic fit explains 37.9% of response variance. The retrieval-window parameter χr = τr pgen/tattempt provides an interpretable descriptor, but an independent-channel approximation does not establish a network memorysizing law. In 5,310 heterogeneous-placement runs, hub and betweenness-based placement outperform peripheral placement in aggregate, with topology-specific exceptions. Coupling assumptions change absolute performance while preserving the aggregate strategy ordering. Harmonized model comparisons provide limited robustness evidence, not1 independent native-protocol validation. The results support model-specific capacity and placement analysis with explicit metric definitions, finite-grid limitations, and interface assumptions

•Oct 13, 2026•0 citations

Improving the Quantum Circuit Problem: A Machine Learning Approach to Gate Depth Prediction Using Linear and Random Forest Regression

Rajendar Dommeti

This study investigates the application of machine learning techniques for predicting gate depth in quantum computational systems, focusing on the relationship between quantum circuit parameters and computational performance. Using a dataset of 50 observations encompassing qubit count, gate depth, entanglement rate, and logical error rate, we developed and compared Linear Regression (LR) and Random Forest Regression (RFR) models for gate depth prediction. The research addresses critical challenges in quantum fault tolerance and circuit optimization, which are essential for achieving scalable quantum computation. Our analysis revealed strong negative correlation (-0.65) between qubit count and logical error rate, while gate depth showed moderate positive correlation (0.62) with error rates. The Random Forest model demonstrated superior performance with R² values of 0.96068 on training data and 0.87262 on testing data, significantly outperforming Linear Regression (R²: 0.87552 training, 0.60257 testing). Error metrics further confirmed RFR's superiority, achieving RMSE of 2.33 (training) and 2.97 (testing) compared to LR's 4.15 and 5.25 respectively. The study highlights the importance of quantum error correction mechanisms and the inherent trade-offs between circuit complexity and computational reliability. Both models showed expected performance degradation from training to testing, with RFR maintaining better generalization capabilities. These findings contribute to the development of typed models for quantum mechanics in computational applications, offering insights into quantum circuit optimization and fault-tolerant design. The research demonstrates that ensemble methods like Random Forest provide more robust predictions for quantum system parameters, supporting the advancement of practical quantum computing implementations and algorithm development. KeyWords: Quantum Computation, Gate Depth Prediction, Random Forest Regression, Quantum Error Correction, Machine Learning, Fault Tolerance.

Journal of Quantum Computing and Advanced Algorithms•Nov 29, 2025•15 citations

Adapting Quantum Machine Learning for Energy Dissociation of Bonds

Swathi Chandrasekhar, Shiva Raj Pokhrel, Navneet Singh

Accurate prediction of bond dissociation energies (BDEs) underpins mechanistic insight and the rational design of molecules and materials. We present a systematic, reproducible benchmark comparing quantum and classical machine learning models for BDE prediction using a chemically curated feature set encompassing atomic properties (atomic numbers, hybridization), bond characteristics (bond order, type), and local environmental descriptors. Our quantum framework, implemented in Qiskit Aer on six qubits, employs ZZFeatureMap encodings with variational ansätz (RealAmplitudes) across multiple architectures Variational Quantum Regressors (VQR), Quantum Support Vector Regressors (QSVR), Quantum Neural Networks (QNN), Quantum Convolutional Neural Networks (QCNN), and Quantum Random Forests (QRF). These are rigorously benchmarked against strong classical baselines, including Support Vector Regression (SVR), Random Forests (RF), and Multi-Layer Perceptrons (MLP). Comprehensive evaluation spanning absolute and relative error metrics, threshold accuracies, and error distributions shows that top-performing quantum models (QCNN, QRF) match the predictive accuracy and robustness of classical ensembles and deep networks, particularly within the chemically prevalent mid-range BDE regime. These findings establish a transparent baseline for quantum-enhanced molecular property prediction and outline a practical foundation for advancing quantum computational chemistry toward near-chemical accuracy.

arXiv:2510.06563•Sep 25, 2025•4 citations

Memory Physics, Finite Capacity, and Topology in Heterogeneous Quantum Networks

David Vesterlund

Quantum-network capacity planning couples memory physics, finite storage, topology, and traffic. We study these interactions with a discrete-event simulator and six representative memory profiles. The simulator is validated against exact single-link, queueing, and swap-ASAP results and against the native SeQUeNCe protocol stack. The native comparison shows that a shared-multiplexing convention used in earlier versions of the model overstates the capacity gain from additional memory, so we report a per-slot model as the primary analysis and the earlier conventions as a sensitivity arm. A single frozen campaign of 266,736 time-based runs with 20 seeds per cell covers an atlas of six profiles, nine topologies, three loads, three network sizes, and seven per-node memory budgets, together with fixed-total-memory, coherence-time, heterogeneous-placement, and Sobol-sensitivity experiments. Under the primary model, the memory budget at which completion reaches 95% of its observed maximum has a median of 8 slots per node, depends more on topology than on memory profile, and reaches the grid boundary of 64 slots in 3.9% of 9688 regimes. Under the earlier conventions, it is strongly profile-dependent, with medians from 4 to at least 64 and 45.8% boundary cases. Four scalar reductions do not collapse the response (pooled quadratic R2 at most 0.152). Topology effects persist at fixed total memory after adjustment for path length. Request completion is independent of the coherence time T2 by construction and is governed by retrieval efficiency, attempt time, and the retrieval decay time τr; T2 enters through fidelity-qualified service. The retrieval-window parameter χr=τrpgen/tattempt orders the profiles by achievable completion but is not a memory-sizing law. Placing high-quality memories at hubs or betweenness bottlenecks outperforms peripheral placement in both model arms and under three interface-coupling rules. Memory-sizing conclusions are therefore sensitive to how multiplexing and storage are modeled and should be validated against native protocol implementations.

Quantum Reports•Nov 10, 2026•0 citations

Can Predictive Models Determine the Optimal Configuration for Quantum Circuits?

William Wallace Rodrigues, João Fabrício Filho

The practical utility of quantum computers in the noisy intermediate scale era depends on error mitigation to preserve computational state fidelity. The search for the ideal compilation configuration, which balances the structural optimization level and the number of statistical shots, represents a problem of high computational cost due to the need for exhaustive empirical evaluations. This work proposes a predictive recommendation system based on Machine Learning to determine the best compilation configuration without requiring prior physical simulations. Using a dataset of 50000 circuits simulated across 28 distinct scenarios, the Decision Tree, Random Forest, and XGBoost models were trained to predict quantum fidelity. The experiments demonstrate that ensemble architectures achieved a Top-3 accuracy of 88.7%, successfully identifying the optimal physical configuration among the 28 possible scenarios in nearly 90% of the cases.

•Aug 12, 2026•0 citations
Findings: On the original quantum-network dataset, the best nonlinear model, Random Forest, reduces mean absolute error (MAE) by only 4.0% relative to Ridge regression with one-hot-encoded categorical features (2.500 versus 2.605), while Ridge reduces MAE by 49.5% relative to a constant predictor (5.160). Nearly all predictive signal is associated with memory family; every individual numerical feature has |r| < 0.08 with the target memory budget mc. In four of five controlled synthetic scenarios—threshold effects, saturation, phase transitions, and family-specific nonlinearity—the best tree ensemble reduces MAE by 31–66% relative to Ridge. In the heterogeneous scenario, Random Forest also reduces the observed under-dimensioning frequency from 55.5% to 35.0%. For a two-way numerical interaction, however, the improvement is only 12.5%. These single-split results support Ridge with one-hot encoding as a necessary baseline, but do not establish a universal model-selection threshold. Method: The original dataset contains 1,000 seed-averaged entanglement-buffering configurations spanning six memory families, nine topology classes, three noise levels, seven traffic levels, and five network sizes (N ϵ {10,15,20,30,50}), with mc ϵ [1,20]. We compare a constant predictor, single-feature linear regression, multivariable linear regression, Ridge regression, Random Forest, and XGBoost on one fixed 60/20/20 train/validation/test split. MAE is the primary metric, supplemented by tolerance accuracy and the fraction of predictions below the target. To probe where nonlinear models begin to add value, we construct five synthetic datasets of 1,000 observations each with controlled threshold, interaction, saturation, phase-transition, and family-heterogeneous response functions. Because all reported comparisons use one split and seed, differences of a few percentage points are exploratory.
Support vector machines are powerful and frequently used machine learning methods for classification and regression tasks, which rely on the construction of kernel matrices. While crucial for the performance of this machine learning approach, choosing the most suitable kernel is highly problem-dependent. The emergence of quantum computers and quantum machine learning techniques provides new possibilities for generating powerful quantum kernels. Within this work, we solve a real-world price forecasting problem using fidelity and projected quantum kernels, which are promising candidates for the utility of near-term quantum computing. In our analysis, we examine and validate the most auspicious quantum kernels from literature and compare their performance with an optimized classical kernel. Unlike previous work on quantum support vector machines, our dataset includes categorical features that need to be encoded as numerical features, which we realize by using the one-hot-encoding scheme. One-hot-encoding, however, increases the dimensionality of the dataset significantly, which collides with the current limitations of noisy intermediate scale quantum computers. To overcome these limitations, we use autoencoders to learn a low-dimensional representation of the feature space that still maintains the most important information of the original data. To examine the impact of autoencoding, we compare the results of the encoded date with the results of the original, unencoded dataset. We could demonstrate that quantum kernels are comparable to or even better than the classical support vector machine kernels regarding the mean absolute percentage error scores for both encoded and unencoded datasets.
Abstract Quantum-network capacity planning couples memory physics, finite storage, topology, and traffic. We study these interactions using a discrete-event simulator with six representative memory profiles. A memory–topology atlas contains 9,360 completed runs, complemented by 8,730 confirmatory runs and an extended per-node memory-budget sweep. We distinguish a threshold relative to the best performance observed on a finite grid from a physical saturation limit. Among 840 reported threshold estimates, 22.3% reach 95% of the observed maximum by four slots per node, while 35.5% first reach it at the upper grid boundary of 64 slots. Four tested scalar reductions do not provide a topology independent collapse; the best pooled quadratic fit explains 37.9% of response variance. The retrieval-window parameter χr = τr pgen/tattempt provides an interpretable descriptor, but an independent-channel approximation does not establish a network memorysizing law. In 5,310 heterogeneous-placement runs, hub and betweenness-based placement outperform peripheral placement in aggregate, with topology-specific exceptions. Coupling assumptions change absolute performance while preserving the aggregate strategy ordering. Harmonized model comparisons provide limited robustness evidence, not1 independent native-protocol validation. The results support model-specific capacity and placement analysis with explicit metric definitions, finite-grid limitations, and interface assumptions
This study investigates the application of machine learning techniques for predicting gate depth in quantum computational systems, focusing on the relationship between quantum circuit parameters and computational performance. Using a dataset of 50 observations encompassing qubit count, gate depth, entanglement rate, and logical error rate, we developed and compared Linear Regression (LR) and Random Forest Regression (RFR) models for gate depth prediction. The research addresses critical challenges in quantum fault tolerance and circuit optimization, which are essential for achieving scalable quantum computation. Our analysis revealed strong negative correlation (-0.65) between qubit count and logical error rate, while gate depth showed moderate positive correlation (0.62) with error rates. The Random Forest model demonstrated superior performance with R² values of 0.96068 on training data and 0.87262 on testing data, significantly outperforming Linear Regression (R²: 0.87552 training, 0.60257 testing). Error metrics further confirmed RFR's superiority, achieving RMSE of 2.33 (training) and 2.97 (testing) compared to LR's 4.15 and 5.25 respectively. The study highlights the importance of quantum error correction mechanisms and the inherent trade-offs between circuit complexity and computational reliability. Both models showed expected performance degradation from training to testing, with RFR maintaining better generalization capabilities. These findings contribute to the development of typed models for quantum mechanics in computational applications, offering insights into quantum circuit optimization and fault-tolerant design. The research demonstrates that ensemble methods like Random Forest provide more robust predictions for quantum system parameters, supporting the advancement of practical quantum computing implementations and algorithm development. KeyWords: Quantum Computation, Gate Depth Prediction, Random Forest Regression, Quantum Error Correction, Machine Learning, Fault Tolerance.
Accurate prediction of bond dissociation energies (BDEs) underpins mechanistic insight and the rational design of molecules and materials. We present a systematic, reproducible benchmark comparing quantum and classical machine learning models for BDE prediction using a chemically curated feature set encompassing atomic properties (atomic numbers, hybridization), bond characteristics (bond order, type), and local environmental descriptors. Our quantum framework, implemented in Qiskit Aer on six qubits, employs ZZFeatureMap encodings with variational ansätz (RealAmplitudes) across multiple architectures Variational Quantum Regressors (VQR), Quantum Support Vector Regressors (QSVR), Quantum Neural Networks (QNN), Quantum Convolutional Neural Networks (QCNN), and Quantum Random Forests (QRF). These are rigorously benchmarked against strong classical baselines, including Support Vector Regression (SVR), Random Forests (RF), and Multi-Layer Perceptrons (MLP). Comprehensive evaluation spanning absolute and relative error metrics, threshold accuracies, and error distributions shows that top-performing quantum models (QCNN, QRF) match the predictive accuracy and robustness of classical ensembles and deep networks, particularly within the chemically prevalent mid-range BDE regime. These findings establish a transparent baseline for quantum-enhanced molecular property prediction and outline a practical foundation for advancing quantum computational chemistry toward near-chemical accuracy.
Quantum-network capacity planning couples memory physics, finite storage, topology, and traffic. We study these interactions with a discrete-event simulator and six representative memory profiles. The simulator is validated against exact single-link, queueing, and swap-ASAP results and against the native SeQUeNCe protocol stack. The native comparison shows that a shared-multiplexing convention used in earlier versions of the model overstates the capacity gain from additional memory, so we report a per-slot model as the primary analysis and the earlier conventions as a sensitivity arm. A single frozen campaign of 266,736 time-based runs with 20 seeds per cell covers an atlas of six profiles, nine topologies, three loads, three network sizes, and seven per-node memory budgets, together with fixed-total-memory, coherence-time, heterogeneous-placement, and Sobol-sensitivity experiments. Under the primary model, the memory budget at which completion reaches 95% of its observed maximum has a median of 8 slots per node, depends more on topology than on memory profile, and reaches the grid boundary of 64 slots in 3.9% of 9688 regimes. Under the earlier conventions, it is strongly profile-dependent, with medians from 4 to at least 64 and 45.8% boundary cases. Four scalar reductions do not collapse the response (pooled quadratic R2 at most 0.152). Topology effects persist at fixed total memory after adjustment for path length. Request completion is independent of the coherence time T2 by construction and is governed by retrieval efficiency, attempt time, and the retrieval decay time τr; T2 enters through fidelity-qualified service. The retrieval-window parameter χr=τrpgen/tattempt orders the profiles by achievable completion but is not a memory-sizing law. Placing high-quality memories at hubs or betweenness bottlenecks outperforms peripheral placement in both model arms and under three interface-coupling rules. Memory-sizing conclusions are therefore sensitive to how multiplexing and storage are modeled and should be validated against native protocol implementations.
The practical utility of quantum computers in the noisy intermediate scale era depends on error mitigation to preserve computational state fidelity. The search for the ideal compilation configuration, which balances the structural optimization level and the number of statistical shots, represents a problem of high computational cost due to the need for exhaustive empirical evaluations. This work proposes a predictive recommendation system based on Machine Learning to determine the best compilation configuration without requiring prior physical simulations. Using a dataset of 50000 circuits simulated across 28 distinct scenarios, the Decision Tree, Random Forest, and XGBoost models were trained to predict quantum fidelity. The experiments demonstrate that ensemble architectures achieved a Top-3 accuracy of 88.7%, successfully identifying the optimal physical configuration among the 28 possible scenarios in nearly 90% of the cases.