A variational autoencoder learns quantum embeddings compressing ImageNet into 13 qubits and achieving 98.5% accuracy on MNIST 3-vs-5 classification with a quantum circuit, close to classical baselines and far above naive amplitude embeddings.
hub Canonical reference
Better than classical? the subtle artofbenchmarkingquantummachinelearningmodels
Canonical reference. 83% of citing Pith papers cite this work as background.
26
Pith papers citing it
Background
83% of classified citations
abstract
Benchmarking models via classical simulations is one of the main ways to judge ideas in quantum machine learning before noise-free hardware is available. However, the huge impact of the experimental design on the results, the small scales within reach today, as well as narratives influenced by the commercialisation of quantum technologies make it difficult to gain robust insights. To facilitate better decision-making we develop an open-source package based on the PennyLane software framework and use it to conduct a large-scale study that systematically tests 12 popular quantum machine learning models on 6 binary classification tasks used to create 160 individual datasets. We find that overall, out-of-the-box classical machine learning models outperform the quantum classifiers. Moreover, removing entanglement from a quantum model often results in as good or better performance, suggesting that "quantumness" may not be the crucial ingredient for the small learning tasks considered here. Our benchmarks also unlock investigations beyond simplistic leaderboard comparisons, and we identify five important questions for quantum model design that follow from our results.
hub tools
citation-role summary
background 6
citation-polarity summary
roles
background 6representative citing papers
QHA represents order-k token interactions in O(log k) quantum circuit depth, with an expressivity separation from classical self-attention and empirical gains on high-order parity and application tasks at reduced parameter count.
MCTS discovers superior data encoding circuits for QCCNNs that outperform standard encodings on medical datasets, with effective rank of feature maps serving as a performance predictor.
QCNN layers equivariant under pixel cyclic shifts are exactly characterized as Fourier-mode multiplexers after QFT, enabling a deep network with constant expected gradient norm at initialization.
A Pauli-transfer-matrix analysis of QELMs reveals the full set of nonlinear Pauli features generated by encoding and transformed by quantum channels, producing an interpretable classical nonlinear vector autoregression model that approximates flow maps in dynamical systems.
MerLin is a new open-source discovery engine for photonic and hybrid quantum machine learning that integrates circuit simulations into standard ML frameworks and reproduces 18 prior works as reusable benchmarks.
Overparameterized two-qubit SU(4) QNNs exhibit grokking and epoch-wise double descent; depth raises generalization success, and weak L2 regularization anchors the post-grokking state against weight-norm drift.
NEB-adapted ravine ensembles for QNNs classifying concentratable entanglement outperform naive methods when local-prediction variability is high and reduce costs, with ravines persisting under depth and qubit scaling.
Split-head QGAN generates more structurally diverse and novel metastable Mg-Mn-O crystals than its classical ablation counterpart but shows lower thermodynamic precision.
Q-ANCHOR combines ZNE-guided server anchoring with stateful client correction to reduce both non-IID client drift and quantum hardware bias in federated quantum learning.
Directly training soft-unitary matrices with a unitarity regularization term and converting them to circuits via alignment enables faster training and lower loss than gate-based optimization on small quantum classification and reinforcement learning tasks.
A variational quantum circuit trained solely on classical measurement outcomes reconstructs diverse quantum states including GHZ, spin-chain ground states, and random circuits with fidelities above 90% on simulators and real NISQ hardware.
Joint multi-expert unitary convolution plus stabilizer-syndrome features roughly halves MNIST/Fashion-MNIST failure rates versus classical diffusion pooling, at moderate GPU overhead.
RP000, a room-temperature CMOS photonic three-qubit processor, delivers higher or comparable accuracy to parameter-matched classical nets on ML classification and better noise tolerance than a superconducting processor.
Compares shadow-based CNNs and physics-informed QCNNs for predicting low-energy subspace overlaps in quenched 10-qubit Heisenberg chains, reporting regime-dependent R^2 performance with QCNNs more stable overall.
A controlled benchmark for context-sensitive memory shows adaptive plasticity (especially homeostatic) enables recall under weak support, with quantum-like models preserving order sensitivity better than Markov controls but without universal advantage.
Quantum kernels in QSVM deliver higher minority-class F1 scores than classical linear or RBF kernels on medical foundation model embeddings for binary insurance classification, avoiding classical collapse in noiseless simulation.
Classical machine learning models outperform variational quantum models on population-level prediction of heat-related physiological events, while quantum models still exhibit non-trivial learning on the harmonized datasets.
Quantum-inspired 1024-D document embeddings exhibit weak, unstable ranking performance and structural geometric limitations, performing better as auxiliary components in hybrid lexical-embedding retrieval systems.
Shot-based quantum encoding embeds data as a classical probability mixture over basis states, making the quantum layer a linear map on probabilities—structurally a classical MLP.
Quantum computers may enable more natural manipulation of Fourier spectra in ML models via the Quantum Fourier Transform, potentially leading to resource-efficient spectral methods.
IQPopt is a JAX-based software tool enabling classical optimization of IQP circuits with thousands of qubits via efficient simulation of Pauli-Z expectation values, plus a module for quantum generative model training.
QCNN, QRNN, and QViT perform well on low-feature data but degrade on high-feature datasets, with QViT most robust to quantum noise and classical-style models better against adversarial noise.
A correlation-guided hybrid quantum-classical model using QAOA achieves 84.6% accuracy on crime pattern classification with reduced trainable parameters compared to classical machine learning baselines.
citing papers explorer
-
Tailor Made Embeddings for Quantum Machine Learning
A variational autoencoder learns quantum embeddings compressing ImageNet into 13 qubits and achieving 98.5% accuracy on MNIST 3-vs-5 classification with a quantum circuit, close to classical baselines and far above naive amplitude embeddings.
-
Higher-Order Token Interactions via Quantum Attention
QHA represents order-k token interactions in O(log k) quantum circuit depth, with an expressivity separation from classical self-attention and empirical gains on high-order parity and application tasks at reduced parameter count.
-
Discovering Data Encoding Strategies for Quantum-Classical Neural Networks Using Monte Carlo Tree Search
MCTS discovers superior data encoding circuits for QCCNNs that outperform standard encodings on medical datasets, with effective rank of feature maps serving as a performance predictor.
-
Pixel-Translation-Equivariant Quantum Convolutional Neural Networks via Fourier Multiplexers
QCNN layers equivariant under pixel cyclic shifts are exactly characterized as Fourier-mode multiplexers after QFT, enabling a deep network with constant expected gradient norm at initialization.
-
Theory and interpretability of Quantum Extreme Learning Machines: a Pauli-transfer matrix approach
A Pauli-transfer-matrix analysis of QELMs reveals the full set of nonlinear Pauli features generated by encoding and transformed by quantum channels, producing an interpretable classical nonlinear vector autoregression model that approximates flow maps in dynamical systems.
-
MerLin: A Discovery Engine for Photonic and Hybrid Quantum Machine Learning
MerLin is a new open-source discovery engine for photonic and hybrid quantum machine learning that integrates circuit simulations into standard ML frameworks and reproduces 18 prior works as reusable benchmarks.
-
Grokking and epoch-wise double descent in quantum neural networks
Overparameterized two-qubit SU(4) QNNs exhibit grokking and epoch-wise double descent; depth raises generalization success, and weak L2 regularization anchors the post-grokking state against weight-norm drift.
-
Ravines in quantum cost landscapes: opportunities for improved VQA predictions
NEB-adapted ravine ensembles for QNNs classifying concentratable entanglement outperform naive methods when local-prediction variability is high and reduce costs, with ravines persisting under depth and qubit scaling.
-
Split-Head Quantum Generative Adversarial Network for Crystalline Material Discovery
Split-head QGAN generates more structurally diverse and novel metastable Mg-Mn-O crystals than its classical ablation counterpart but shows lower thermodynamic precision.
-
Q-ANCHOR: Federated Quantum Learning with ZNE-guided Correction
Q-ANCHOR combines ZNE-guided server anchoring with stateful client correction to reduce both non-IID client drift and quantum hardware bias in federated quantum learning.
-
Soft-Quantum Algorithms
Directly training soft-unitary matrices with a unitarity regularization term and converting them to circuits via alignment enables faster training and lower loss than gate-based optimization on small quantum classification and reinforcement learning tasks.
-
Quantum Machine Learning for State Tomography Using Classical Data
A variational quantum circuit trained solely on classical measurement outcomes reconstructs diverse quantum states including GHZ, spin-chain ground states, and random circuits with fidelities above 90% on simulators and real NISQ hardware.
-
Image classification via a quantum-inspired strategy involving a mixture of experts
Joint multi-expert unitary convolution plus stabilizer-syndrome features roughly halves MNIST/Fashion-MNIST failure rates versus classical diffusion pooling, at moderate GPU overhead.
-
Design and Benchmarking of a Quantum Photonic Chip
RP000, a room-temperature CMOS photonic three-qubit processor, delivers higher or comparable accuracy to parameter-matched classical nets on ML classification and better noise tolerance than a superconducting processor.
-
Learning Low-Energy Subspace Overlaps in Many-Body Systems with Measurement-Based and Coherent Quantum Strategies
Compares shadow-based CNNs and physics-informed QCNNs for predicting low-energy subspace overlaps in quenched 10-qubit Heisenberg chains, reporting regime-dependent R^2 performance with QCNNs more stable overall.
-
A quantum-like benchmark for context-sensitive associative memory with adaptive plasticity
A controlled benchmark for context-sensitive memory shows adaptive plasticity (especially homeostatic) enables recall under weak support, with quantum-like models preserving order sensitivity better than Markov controls but without universal advantage.
-
Quantum Kernel Advantage over Classical Collapse in Medical Foundation Model Embeddings
Quantum kernels in QSVM deliver higher minority-class F1 scores than classical linear or RBF kernels on medical foundation model embeddings for binary insurance classification, avoiding classical collapse in noiseless simulation.
-
Classical and Quantum Machine Learning for Population-Level Prediction of Heat-Related Physiological Events
Classical machine learning models outperform variational quantum models on population-level prediction of heat-related physiological events, while quantum models still exhibit non-trivial learning on the harmonized datasets.
-
On the Representational Limits of Quantum-Inspired 1024-D Document Embeddings: An Experimental Evaluation Framework
Quantum-inspired 1024-D document embeddings exhibit weak, unstable ranking performance and structural geometric limitations, performing better as auxiliary components in hybrid lexical-embedding retrieval systems.
-
Shot-based quantum encoding: a data-loading paradigm for quantum neural networks
Shot-based quantum encoding embeds data as a classical probability mixture over basis states, making the quantum layer a linear map on probabilities—structurally a classical MLP.
-
Spectral methods: crucial for machine learning, natural for quantum computers?
Quantum computers may enable more natural manipulation of Fourier spectra in ML models via the Quantum Fourier Transform, potentially leading to resource-efficient spectral methods.
-
IQPopt: Fast optimization of instantaneous quantum polynomial circuits in JAX
IQPopt is a JAX-based software tool enabling classical optimization of IQP circuits with thousands of qubits via efficient simulation of Pauli-Z expectation values, plus a module for quantum generative model training.
-
A Comprehensive Analysis of Accuracy and Robustness in Quantum Neural Networks
QCNN, QRNN, and QViT perform well on low-feature data but degrade on high-feature datasets, with QViT most robust to quantum noise and classical-style models better against adversarial noise.
-
Domain-Aware Hybrid Quantum Learning via Correlation-Guided Circuit Design for Crime Pattern Analytics
A correlation-guided hybrid quantum-classical model using QAOA achieves 84.6% accuracy on crime pattern classification with reduced trainable parameters compared to classical machine learning baselines.
-
When AI meets quantum information: A comprehensive review
A comprehensive review organizing progress at the AI-quantum information intersection from both directions.
-
Quantum in Biology, Quantum for Biology, and Biology for Quantum: Mapping the Evidence and the Road Ahead
A narrative evidence map of quantum-biology intersections that evaluates mechanistic claims, invoked quantum resources, key experiments, classical confounds, and decisive benchmarks for each of three complementary directions.