[
  {
    "assumptions": [
      "Output-layer-only training with fixed reservoir dynamics."
    ],
    "citation": "S01 (2025)",
    "claims": [
      "Entanglement-correlated regimes improve embedding quality without proving asymptotic quantum advantage."
    ],
    "conclusions": [
      "Representational gains can coexist with classically simulable operating points."
    ],
    "contributions": [
      "Quantifies accuracy transition versus evolution time on image tasks."
    ],
    "future_work": [
      "Test harder datasets and deeper/nonintegrable dynamics."
    ],
    "key_equations": [
      "H=(1/2)\u2211i(\u03c3x(i)\u03c3x(i+1)+\u03c3y(i)\u03c3y(i+1))"
    ],
    "limitations": [
      "Moderate system sizes and benchmark-scale datasets."
    ],
    "source_type": "paper",
    "summary": "QELM image-study linking an entanglement onset regime to improved separability while noting broad classical simulability.",
    "title": "Entanglement and Classical Simulability in Quantum Extreme Learning Machines",
    "url": "https://arxiv.org/abs/2509.06873",
    "year": 2025
  },
  {
    "assumptions": [
      "PCA/AE compression preserves task signal."
    ],
    "citation": "S02 (2025)",
    "claims": [
      "Encoding choice strongly affects final accuracy."
    ],
    "conclusions": [
      "Preprocessing, encoding, and measurement must be co-designed."
    ],
    "contributions": [
      "Compares encoding and Hamiltonian families under matched readouts."
    ],
    "future_work": [
      "Hardware execution with calibrated measurement constraints."
    ],
    "key_equations": [
      "|\u03c8\u27e9=cos(\u03b8/2)|0\u27e9+ei\u03c6sin(\u03b8/2)|1\u27e9"
    ],
    "limitations": [
      "Primarily simulation-oriented assessment."
    ],
    "source_type": "paper",
    "summary": "End-to-end QELM image pipeline with encoding and Hamiltonian ablations.",
    "title": "Harnessing Quantum Extreme Learning Machines for image classification",
    "url": "https://arxiv.org/abs/2409.00998",
    "year": 2025
  },
  {
    "assumptions": [
      "Fixed reservoir map, optimized readout dominates gains."
    ],
    "citation": "S03 (2026)",
    "claims": [
      "Kernel-space optimization improves efficiency and accuracy versus naive readout search."
    ],
    "conclusions": [
      "Measurement design is a primary practical lever."
    ],
    "contributions": [
      "Derives dual/kernel readout optimization for QELM and recurrent QRC."
    ],
    "future_work": [
      "Add hardware-aware regularization and non-MSE losses."
    ],
    "key_equations": [
      "K(x,x\u2032)=tr(\u03c1(x)\u03c1(x\u2032))"
    ],
    "limitations": [
      "Hardware-observable constraints remain."
    ],
    "source_type": "paper",
    "summary": "Casts QRC observable optimization as kernel regression for fixed reservoirs.",
    "title": "Kernel-based optimization of measurement operators for quantum reservoir computers",
    "url": "https://arxiv.org/abs/2602.14677",
    "year": 2026
  },
  {
    "assumptions": [
      "Data encoding implemented via exp(-ixH)-style operators."
    ],
    "citation": "S04 (2021)",
    "claims": [
      "Encoding is a first-order determinant of model expressivity."
    ],
    "conclusions": [
      "Feature-map design is central to QML performance."
    ],
    "contributions": [
      "Formal expressive-power characterization through encoding-induced frequencies."
    ],
    "future_work": [
      "Finite-sample/noisy-device expressivity studies."
    ],
    "key_equations": [
      "f\u03b8(x)=\u2211\u03c9\u2208\u03a9 c\u03c9(\u03b8)e^{i\u03c9x}"
    ],
    "limitations": [
      "Theory-heavy; limited hardware-noise treatment."
    ],
    "source_type": "paper",
    "summary": "Fourier-spectrum analysis showing encoding operators determine reachable function classes.",
    "title": "The effect of data encoding on the expressive power of variational quantum machine learning models",
    "url": "https://arxiv.org/abs/2008.08605",
    "year": 2021
  },
  {
    "assumptions": [
      "Quantum models are state-preparation plus measurement pipelines."
    ],
    "citation": "S05 (2021)",
    "claims": [
      "Data encoding largely determines inductive bias via kernel choice."
    ],
    "conclusions": [
      "Kernel view gives clearer optimization and baseline design."
    ],
    "contributions": [
      "Operator-space/kernel interpretation for supervised QML."
    ],
    "future_work": [
      "Scalable kernel-estimation and complexity-separation studies."
    ],
    "key_equations": [
      "\u03ba(x,x\u2032)=tr(\u03c1(x)\u03c1(x\u2032))"
    ],
    "limitations": [
      "Kernel computation can be costly at scale."
    ],
    "source_type": "paper",
    "summary": "Unifies supervised QML with kernel methods and optimal measurement expansions.",
    "title": "Supervised quantum machine learning models are kernel methods",
    "url": "https://arxiv.org/abs/2101.11020",
    "year": 2021
  },
  {
    "assumptions": [
      "Finite-shot statistics govern practical performance."
    ],
    "citation": "S06 (2024)",
    "claims": [
      "Competitive performance is feasible but shot-limited."
    ],
    "conclusions": [
      "Measurement budget is a core deployment bottleneck."
    ],
    "contributions": [
      "Characterizes shot-budget and runtime-performance tradeoffs."
    ],
    "future_work": [
      "Larger arrays and stronger multiclass protocols."
    ],
    "key_equations": [
      "y=Wout\u00b7z where z are measured reservoir features"
    ],
    "limitations": [
      "Benchmark scale and hardware drift constraints."
    ],
    "source_type": "paper",
    "summary": "Neutral-atom QRC workflow with finite-shot tradeoff analysis on practical tasks.",
    "title": "Practical Quantum Reservoir Computing in Rydberg Atom Arrays",
    "url": "https://arxiv.org/abs/2407.02553",
    "year": 2024
  },
  {
    "assumptions": [
      "Feature selection and reduced-dimensional encodings retain chemical signal."
    ],
    "citation": "S07 (2024)",
    "claims": [
      "QRC features degrade more slowly as training size decreases."
    ],
    "conclusions": [
      "Reservoir embeddings can help small/noisy scientific datasets."
    ],
    "contributions": [
      "Integrates SHAP feature filtering with QRC embedding and downstream regressors."
    ],
    "future_work": [
      "Broader molecular benchmarks and hardware trials."
    ],
    "key_equations": [
      "H(t)=\u03a9/2\u2211j(|gj\u27e9\u27e8rj|+|rj\u27e9\u27e8gj|)+\u2211j<kVjknjnk-\u2211j\u0394jnj"
    ],
    "limitations": [
      "Simulation-centric and dataset-specific scope."
    ],
    "source_type": "paper",
    "summary": "QRC embeddings for molecular prediction with robustness in low-data regimes.",
    "title": "Robust Quantum Reservoir Computing for Molecular Property Prediction",
    "url": "https://arxiv.org/abs/2412.06758",
    "year": 2024
  },
  {
    "assumptions": [
      "Fixed quantum dynamics can serve as reservoir substrate."
    ],
    "citation": "S08 (2022)",
    "claims": [
      "Rydberg arrays are viable for QRC temporal/inference tasks."
    ],
    "conclusions": [
      "Neutral-atom hardware is a practical QRC candidate."
    ],
    "contributions": [
      "Establishes near-term neutral-atom QRC formulation."
    ],
    "future_work": [
      "Larger devices and stronger baseline-matched studies."
    ],
    "key_equations": [
      "f(x)=wT\u03c6(x) with \u03c6 from fixed quantum evolution and measurement"
    ],
    "limitations": [
      "Near-term scale and noise constraints."
    ],
    "source_type": "paper",
    "summary": "Foundational Rydberg-array QRC architecture and task demonstrations.",
    "title": "Quantum Reservoir Computing Using Arrays of Rydberg Atoms",
    "url": "https://doi.org/10.1103/PRXQuantum.3.030325",
    "year": 2022
  },
  {
    "assumptions": [
      "Measurement feedback can restore erased temporal information."
    ],
    "citation": "S09 (2024)",
    "claims": [
      "Feedback mitigates memory loss from destructive measurement."
    ],
    "conclusions": [
      "Feedback is an effective memory mechanism in measured QRC."
    ],
    "contributions": [
      "Defines and evaluates feedback-driven QRC for time-series tasks."
    ],
    "future_work": [
      "Optimize feedback protocols for hardware-limited settings."
    ],
    "key_equations": [
      "xt+1=F(xt,ut,mt) where mt is fed-back measurement outcome"
    ],
    "limitations": [
      "Performance depends on feedback policy and task regime."
    ],
    "source_type": "paper",
    "summary": "Introduces feedback to recover temporal memory after projective measurement in QRC.",
    "title": "Feedback-driven quantum reservoir computing for time-series analysis",
    "url": "https://doi.org/10.1103/PRXQuantum.5.040325",
    "year": 2024
  },
  {
    "assumptions": [
      "Photonic measurements provide stable reservoir feature streams."
    ],
    "citation": "S10 (2023)",
    "claims": [
      "Photonic architecture supports high-throughput reservoir inference."
    ],
    "conclusions": [
      "Photonics is promising for low-latency QRC deployment."
    ],
    "contributions": [
      "Real-time photonic QRC implementation with scaling analysis."
    ],
    "future_work": [
      "Cross-domain benchmarks and robust calibration transfer."
    ],
    "key_equations": [
      "y=Wout\u00b7[\u27e8n\u27e9,Cov(n)]"
    ],
    "limitations": [
      "Task specialization and platform-specific calibration."
    ],
    "source_type": "paper",
    "summary": "Photonic QRC platform focused on throughput and scalability.",
    "title": "Scalable photonic platform for real-time quantum reservoir computing",
    "url": "https://doi.org/10.1103/PhysRevApplied.20.014051",
    "year": 2023
  },
  {
    "assumptions": [
      "Finite-dimensional reservoirs induce bounded memory/expressivity tradeoffs."
    ],
    "citation": "S11 (2023)",
    "claims": [
      "Reservoir dimension strongly limits achievable memory structure."
    ],
    "conclusions": [
      "Capacity planning is required for fair QRC comparisons."
    ],
    "contributions": [
      "Formal constraints for finite-dimensional QRC behavior."
    ],
    "future_work": [
      "Empirical validation across physical platforms."
    ],
    "key_equations": [
      "MC=\u2211k MCk"
    ],
    "limitations": [
      "Theoretical abstractions may omit hardware-specific effects."
    ],
    "source_type": "paper",
    "summary": "Finite-dimensional analysis of QRC memory and expressivity constraints.",
    "title": "Quantum reservoir computing in finite dimensions",
    "url": "https://doi.org/10.1103/PhysRevE.107.035306",
    "year": 2023
  },
  {
    "assumptions": [
      "Atomic-lattice dynamics provide useful nonlinear features."
    ],
    "citation": "S12 (2025)",
    "claims": [
      "Atomic-lattice reservoirs can achieve competitive predictive performance."
    ],
    "conclusions": [
      "Lattice structure is a key design factor."
    ],
    "contributions": [
      "Joint evaluation on classification and time-series tasks."
    ],
    "future_work": [
      "Topology-aware optimization with stronger baselines."
    ],
    "key_equations": [
      "yt=Wout\u00b7zt"
    ],
    "limitations": [
      "Benchmark and hyperparameter dependence."
    ],
    "source_type": "paper",
    "summary": "Atomic-lattice QRC for classification and forecasting under practical constraints.",
    "title": "Quantum reservoir computing in atomic lattices for classification and time series prediction",
    "url": "https://doi.org/10.1016/j.chaos.2025.116289",
    "year": 2025
  },
  {
    "assumptions": [
      "Scalability depends on measurement/readout-efficient designs."
    ],
    "citation": "S13 (2024)",
    "claims": [
      "Careful readout design improves scalability envelope."
    ],
    "conclusions": [
      "System-level tradeoffs dominate near-term QRC viability."
    ],
    "contributions": [
      "Engineering guidance for practical QRC scaling."
    ],
    "future_work": [
      "Unified benchmark protocol across hardware families."
    ],
    "key_equations": [
      "yt=Wout\u00b7\u03c6(xt)"
    ],
    "limitations": [
      "Platform-specific assumptions."
    ],
    "source_type": "paper",
    "summary": "Practical design rules for scalable QRC implementations.",
    "title": "Practical and Scalable Quantum Reservoir Computing",
    "url": "https://arxiv.org/abs/2405.04799",
    "year": 2024
  },
  {
    "assumptions": [
      "Feature-map symmetry/connectivity governs generalization."
    ],
    "citation": "S14 (2025)",
    "claims": [
      "Simpler topologies can outperform highly connected alternatives."
    ],
    "conclusions": [
      "Connectivity should be tuned to task geometry."
    ],
    "contributions": [
      "Links XY connectivity patterns to QERC performance."
    ],
    "future_work": [
      "Transfer analysis across hardware-native couplings."
    ],
    "key_equations": [
      "HXY=\u2211ij Jij(\u03c3x_i\u03c3x_j+\u03c3y_i\u03c3y_j)"
    ],
    "limitations": [
      "Model-family-specific conclusions."
    ],
    "source_type": "paper",
    "summary": "Analyzes XY-model feature maps and connectivity effects in QERC.",
    "title": "Effective quantum feature maps in quantum extreme reservoir computation from the XY model",
    "url": "https://doi.org/10.1103/PhysRevA.111.022431",
    "year": 2025
  },
  {
    "assumptions": [
      "Classical ESP is insufficient for non-stationary QRC."
    ],
    "citation": "S15 (2024)",
    "claims": [
      "Extended ESP aligns better with observed memory behavior."
    ],
    "conclusions": [
      "ESP generalization improves theoretical diagnostics for QRC."
    ],
    "contributions": [
      "Defines non-stationary and subset/subspace ESP for QRC."
    ],
    "future_work": [
      "Apply ESP diagnostics on hardware-resolved noise models."
    ],
    "key_equations": [
      "xt=(ut,ut-1,\u2026,ut-L)"
    ],
    "limitations": [
      "Benchmark-focused validation."
    ],
    "source_type": "paper",
    "summary": "Extends ESP definitions to non-stationary/subspace QRC regimes.",
    "title": "Extending echo state property for quantum reservoir computing",
    "url": "https://arxiv.org/abs/2403.02686",
    "year": 2024
  },
  {
    "assumptions": [
      "PTM spectral characteristics reflect memory stability."
    ],
    "citation": "S16 (2024)",
    "claims": [
      "Hierarchical criteria explain differing memory regimes."
    ],
    "conclusions": [
      "PTM-based diagnostics are practical for QRC analysis."
    ],
    "contributions": [
      "ESP hierarchy for finer-grained QRC stability analysis."
    ],
    "future_work": [
      "Validate hierarchy with large-scale experimental traces."
    ],
    "key_equations": [
      "\u03c1(PTM)<1 as fading-memory indicator"
    ],
    "limitations": [
      "Theory-led; limited broad hardware validation."
    ],
    "source_type": "paper",
    "summary": "Proposes hierarchy-style ESP criteria and memory diagnostics for QRC.",
    "title": "Hierarchy of the echo state property in quantum reservoir computing",
    "url": "https://arxiv.org/abs/2403.03508",
    "year": 2024
  },
  {
    "assumptions": [
      "Graph topology controls feature diversity and memory."
    ],
    "citation": "S17 (2025)",
    "claims": [
      "Topology materially changes QRC task performance."
    ],
    "conclusions": [
      "Graph design is an optimization variable, not a fixed choice."
    ],
    "contributions": [
      "Systematic random-regular-graph topology ablations."
    ],
    "future_work": [
      "Topology search under hardware constraints."
    ],
    "key_equations": [
      "yt=Wout\u00b7\u03c6G(xt)"
    ],
    "limitations": [
      "Topology-space coverage is finite."
    ],
    "source_type": "paper",
    "summary": "Graph-topology study of QRC stability and prediction quality.",
    "title": "Quantum reservoir computing on random regular graphs",
    "url": "https://doi.org/10.1103/PhysRevA.112.012622",
    "year": 2025
  },
  {
    "assumptions": [
      "Controlled noise can increase useful feature diversity."
    ],
    "citation": "S18 (2023)",
    "claims": [
      "Proper noise settings improve forecasting accuracy."
    ],
    "conclusions": [
      "Noise can be a design knob, not only a nuisance."
    ],
    "contributions": [
      "Noise-induced QRC tuning framework on chaotic tasks."
    ],
    "future_work": [
      "Robust noise schedules under hardware drift."
    ],
    "key_equations": [
      "xt+1=F\u03b8(xt,ut,\u03b7)"
    ],
    "limitations": [
      "Task-specific tuning sensitivity."
    ],
    "source_type": "paper",
    "summary": "Shows noise parameterization can improve nonlinear/chaotic forecasting performance.",
    "title": "Optimizing quantum noise-induced reservoir computing for nonlinear and chaotic time series prediction",
    "url": "https://arxiv.org/abs/2303.05488",
    "year": 2023
  },
  {
    "assumptions": [
      "Repeated measurements can extract richer features."
    ],
    "citation": "S19 (2023)",
    "claims": [
      "Repeated measurement improves practical utility of QRC features."
    ],
    "conclusions": [
      "Measurement protocol design strongly affects outcome quality."
    ],
    "contributions": [
      "Superconducting QRC protocol with repeated readout."
    ],
    "future_work": [
      "Optimize shot allocation and adaptive measurements."
    ],
    "key_equations": [
      "zt=[m1,\u2026,mk] from repeated projective outcomes"
    ],
    "limitations": [
      "Hardware-specific overhead and calibration needs."
    ],
    "source_type": "paper",
    "summary": "Repeated-measurement protocol for practical QRC on superconducting systems.",
    "title": "Quantum reservoir with repeated measurements on superconducting devices",
    "url": "https://arxiv.org/abs/2310.06706",
    "year": 2023
  },
  {
    "assumptions": [
      "Analog dynamics provide stable high-rate reservoir features."
    ],
    "citation": "S20 (2023)",
    "claims": [
      "Analog QRC can achieve competitive task-level performance."
    ],
    "conclusions": [
      "QRC is viable beyond toy software benchmarks."
    ],
    "contributions": [
      "Hardware-aligned analog QRC workflow for RF processing."
    ],
    "future_work": [
      "Latency/energy benchmarking across domains."
    ],
    "key_equations": [
      "yt=Wout\u00b7zt"
    ],
    "limitations": [
      "Domain-specific evaluation scope."
    ],
    "source_type": "paper",
    "summary": "Analog QRC demonstration for microwave signal-processing tasks.",
    "title": "Microwave signal processing using an analog quantum reservoir computer",
    "url": "https://arxiv.org/abs/2312.16166",
    "year": 2023
  },
  {
    "assumptions": [
      "GBS statistics can serve as fixed reservoir features."
    ],
    "citation": "S21 (2025)",
    "claims": [
      "Correlation features materially improve accuracy on reported tasks."
    ],
    "conclusions": [
      "Photonic correlations are useful practical QRC signals."
    ],
    "contributions": [
      "Demonstrates large-scale photonic reservoir feature extraction."
    ],
    "future_work": [
      "Harder multimodal tasks and stronger baseline parity."
    ],
    "key_equations": [
      "features={\u27e8ni\u27e9,Cov(ni,nj)}"
    ],
    "limitations": [
      "Benchmark-scale tasks and classical readout dependence."
    ],
    "source_type": "paper",
    "summary": "Large-scale photonic QRC with >400-mode GBS and correlation-feature gains.",
    "title": "Large-scale quantum reservoir computing using a Gaussian Boson Sampler",
    "url": "https://arxiv.org/abs/2505.13695",
    "year": 2025
  },
  {
    "assumptions": [
      "Finite-shot statistics can induce concentration bottlenecks."
    ],
    "citation": "S22 (2025)",
    "claims": [
      "Hamiltonian symmetries can suppress concentration effects."
    ],
    "conclusions": [
      "Sample complexity is a central scaling bottleneck."
    ],
    "contributions": [
      "Characterizes concentration risk and symmetry mitigation strategy."
    ],
    "future_work": [
      "Platform-specific concentration diagnostics and mitigation."
    ],
    "key_equations": [
      "Var[feature]\u21920 exponentially under concentration regimes"
    ],
    "limitations": [
      "Model assumptions may not cover all hardware regimes."
    ],
    "source_type": "paper",
    "summary": "Analyzes concentration effects and symmetry-based mitigation in QRC sample complexity.",
    "title": "Exponential concentration and symmetries in Quantum Reservoir Computing",
    "url": "https://arxiv.org/abs/2505.10062",
    "year": 2025
  },
  {
    "assumptions": [
      "Chaos diagnostics transfer to QRC design choices."
    ],
    "citation": "S23 (2025)",
    "claims": [
      "Two characteristic edges correlate with improved QRC performance."
    ],
    "conclusions": [
      "Chaos diagnostics can guide operating-point selection."
    ],
    "contributions": [
      "Introduces quantum edge-of-chaos operating-point analysis."
    ],
    "future_work": [
      "Validate criteria on experimentally accessible reservoirs."
    ],
    "key_equations": [
      "performance peak near Thouless-time related boundary"
    ],
    "limitations": [
      "SYK-centered analysis scope."
    ],
    "source_type": "paper",
    "summary": "Identifies QRC performance peaks near many-body chaos boundaries.",
    "title": "Edge of Many-Body Quantum Chaos in Quantum Reservoir Computing",
    "url": "https://arxiv.org/abs/2506.17547",
    "year": 2025
  },
  {
    "assumptions": [
      "PTM is suitable for open-system memory diagnostics."
    ],
    "citation": "S24 (2024)",
    "claims": [
      "Coherence influx is necessary for non-stationary ESP in studied settings."
    ],
    "conclusions": [
      "Open-system coherence flow is a key QRC design variable."
    ],
    "contributions": [
      "Derives non-stationary ESP conditions tied to coherence influx."
    ],
    "future_work": [
      "Empirical PTM diagnostics on noisy hardware traces."
    ],
    "key_equations": [
      "\u03c1(PTM) characterizes fading memory"
    ],
    "limitations": [
      "Theory-heavy with limited hardware calibration detail."
    ],
    "source_type": "paper",
    "summary": "PTM-based analysis showing coherence influx and spectral radius govern non-stationary ESP behavior.",
    "title": "Coherence influx is indispensable for quantum reservoir computing",
    "url": "https://arxiv.org/abs/2409.12693",
    "year": 2024
  },
  {
    "citation": "S25 (2018)",
    "source_type": "article",
    "summary": "Baseline QML framework used to contextualize output-layer-only QRC/QELM alternatives.",
    "title": "Quantum circuit learning",
    "url": "https://doi.org/10.1103/PhysRevA.98.032309",
    "year": 2018
  },
  {
    "citation": "S26 (2019)",
    "source_type": "article",
    "summary": "Feature-space perspective used for kernel-baseline framing.",
    "title": "Quantum machine learning in feature Hilbert spaces",
    "url": "https://doi.org/10.1103/PhysRevLett.122.040504",
    "year": 2019
  },
  {
    "citation": "S27 (2019)",
    "source_type": "article",
    "summary": "Quantum-kernel benchmark reference for matched baseline comparisons.",
    "title": "Supervised learning with quantum-enhanced feature spaces",
    "url": "https://doi.org/10.1038/s41586-019-0980-2",
    "year": 2019
  },
  {
    "citation": "S28 (2021)",
    "source_type": "report",
    "summary": "Survey context for trainability and optimization tradeoffs in quantum ML.",
    "title": "Variational quantum algorithms",
    "url": "https://doi.org/10.1038/s42254-021-00348-9",
    "year": 2021
  },
  {
    "citation": "S29 (2018)",
    "source_type": "article",
    "summary": "Trainability limitation reference motivating fixed-reservoir output-only learning.",
    "title": "Barren plateaus in quantum neural network training landscapes",
    "url": "https://doi.org/10.1038/s41467-018-07090-4",
    "year": 2018
  },
  {
    "citation": "S30 (2001)",
    "source_type": "report",
    "summary": "Foundational echo-state framework used in RC/QRC methodology.",
    "title": "The echo state approach to analysing and training recurrent neural networks",
    "url": "https://www.ai.rug.nl/minds/uploads/EchoStatesTechRep.pdf",
    "year": 2001
  },
  {
    "citation": "S31 (2002)",
    "source_type": "article",
    "summary": "Liquid-state computing foundation for reservoir-style computation.",
    "title": "Real-time computing without stable states",
    "url": "https://doi.org/10.1162/089976602760407955",
    "year": 2002
  },
  {
    "citation": "S32 (2009)",
    "source_type": "report",
    "summary": "Classical RC survey for baseline construction and evaluation norms.",
    "title": "Reservoir computing approaches to recurrent neural network training",
    "url": "https://doi.org/10.1016/j.cosrev.2009.03.005",
    "year": 2009
  },
  {
    "citation": "S33 (1998)",
    "source_type": "dataset",
    "summary": "Canonical image benchmark reference used in many QRC/QELM studies.",
    "title": "MNIST dataset reference",
    "url": "https://doi.org/10.1109/5.726791",
    "year": 1998
  },
  {
    "citation": "S34 (2017)",
    "source_type": "dataset",
    "summary": "Harder drop-in MNIST replacement for benchmark robustness checks.",
    "title": "Fashion-MNIST",
    "url": "https://arxiv.org/abs/1708.07747",
    "year": 2017
  },
  {
    "citation": "S35 (2009)",
    "source_type": "dataset",
    "summary": "Higher-complexity image benchmark for stress-testing representations.",
    "title": "CIFAR-10",
    "url": "https://www.cs.toronto.edu/~kriz/cifar.html",
    "year": 2009
  }
]