Pith. sign in

REVIEW 4 major objections 6 minor 153 references

For the image benchmarks studied, a Quantum Extreme Learning Machine reaches its full accuracy once local entanglement has built up over a finite range, at a timescale independent of system size; from then on its performance equals that of

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 04:01 UTC pith:JQUEJZ4I

load-bearing objection A coherent thesis compiling two published QELM papers; the new material is a preliminary QELM-on-neutrino pilot, and the core claim—local-entanglement plateau that matches Haar-random—is plausible but needs reproducibility and stronger baselines. the 4 major comments →

arxiv 2607.13699 v1 pith:JQUEJZ4I submitted 2026-07-15 quant-ph

Machine learning development for quantum computing and neutrino physics

Annalisa De Lorenzis This is my paper
classification quant-ph
keywords Quantum Extreme Learning MachineXX spin chainentanglement dynamicsclassical simulabilitytensor networksLieb-Robinson boundimage classificationwater Cherenkov detectors
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The thesis tries to establish that the useful operating point of a Quantum Extreme Learning Machine (QELM) for image classification is not a deeply scrambled quantum state but a locally entangled one. In a fixed QELM with dense-angle encoding and an XX spin-chain reservoir, classification accuracy jumps sharply at a characteristic evolution time t* of order 1, essentially independent of the number of qubits, and then sits at a plateau equal to Haar-random unitary processing. Because t* coincides with the Lieb-Robinson light-cone scale, the regime where the QELM works is one of short-range entanglement and finite information spread, compatible with efficient tensor-network simulation. A second, independent contribution shows a modified ResNet-18 classifies single-vertex versus pile-up events in simulated water Cherenkov detector images with AUC close to unity, with timing information especially useful. Taken together, the QELM results reframe reported accuracies not as evidence of generic quantum advantage but as a signature of a classically tractable intermediate regime.

Core claim

The central claim of the QELM part is that the quality of the features generated by the quantum reservoir is governed by the local dynamical timescale of the XX Hamiltonian, not by global entanglement. Across MNIST, Fashion-MNIST, and CIFAR-10, accuracy curves as a function of evolution time show a rapid rise at t* of order 1 for all system sizes N = 6 to 11, and the saturation accuracy matches Haar-random unitaries acting on the same encoded states. The thesis identifies t* with the build-up of single-qubit entropy and the Lieb-Robinson velocity vLR = 1, so that at the operating point correlations have spread only over a finite range; this is exactly the regime where tensor-network simulati

What carries the argument

The machinery is a four-stage pipeline: classical compression (PCA or autoencoder), dense-angle encoding of latent features into an N-qubit product state, fixed unitary evolution under a nearest-neighbour XX Hamiltonian with periodic boundary conditions, and measurement in the computational basis to produce a 2^N-dimensional probability vector for a one-layer neural network readout. The diagnostic objects are Lieb-Robinson bounds, single-qubit and half-chain von Neumann entropies, K-means inertia of the probability polytope, and Haar-random unitaries as benchmarks. The XX chain's Lieb-Robinson velocity vLR = 1 sets the light cone that explains why t* is independent of N and why the plateau s

Load-bearing premise

The load-bearing premise is that the accuracy transition at t* of order 1 is caused by local entanglement growth in the XX chain, rather than by the capacity of the linear readout or by the specific Bloch-angle distribution used in dense-angle encoding; if that attribution fails, the claim that QELMs operate in a locally entangled, classically simulable regime is unsupported.

What would settle it

Fix N and the readout, then vary the distribution of encoded Bloch angles (e.g., uniform versus clustered features) and track whether t* shifts; or compute the tensor-network bond dimension required to reproduce the plateau measurement probabilities at t of order 1 for growing N. If t* moves with encoding or readout choices, or if the required bond dimension grows significantly with N, the local-dynamics and classical-simulability claim would be refuted.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • QELM accuracy curves transition at a system-size-independent t* of order 1 and plateau at Haar-random levels, so evolving beyond t* does not improve classification.
  • The useful regime has only local entanglement and finite information spread, so tensor-network simulation methods can reproduce QELM outputs at the operating point.
  • Local observables such as single-qubit magnetizations lose class information with time while full probability vectors gain it, showing the readout must use global measurement statistics.
  • The IWCD ResNet-18 classifier reaches AUC close to unity for single-vertex versus pile-up on simulated data, with timing channels carrying strong discriminating signal.
  • QELMs can be transferred to realistic neutrino-image data after strong compression, but their performance remains below the classical ResNet baseline.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If t* is robust, a practical design rule for QELMs would be to evolve only for a time of order 1/vLR, cutting circuit depth and noise exposure without sacrificing accuracy.
  • The equality of XX plateau accuracy with Haar-random baselines suggests saturation accuracy may be predictable from the encoding map alone, letting classical simulations set performance ceilings before hardware runs.
  • Because the useful regime is shallow, the XX reservoir could be replaced by shallow random two-qubit circuits; if accuracies match, the QELM's quantum contribution is essentially reproducible with classically simulable circuits.
  • For neutrino classification, the near-unity AUC on simulations should be re-tested on data with controlled vertex separation and positions near detector boundaries, since the appendix's 2D studies show performance degrades for close vertices and truncated rings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This PhD thesis combines two research threads. The first (Chapters 4–6) studies Quantum Extreme Learning Machines for image classification. After reviewing ML and quantum-computing background, it presents a systematic QELM pipeline: classical feature reduction (PCA or autoencoders), data encoding, fixed Hamiltonian evolution, computational-basis measurement, and a one-layer classical readout. On MNIST/Fashion-MNIST it reports that nonlinear autoencoders outperform PCA, that dense-angle and uniform Bloch-sphere encodings are best, and that all sufficiently interacting Hamiltonians perform similarly once N is large. The second QELM study fixes an XX-chain Hamiltonian and varies only the evolution time t. The main claimed results are: (i) accuracy vs t shows a sharp transition at a characteristic t* ≈ 1 that is essentially independent of N = 6–11 on MNIST, Fashion-MNIST, and CIFAR-10; (ii) the plateau accuracy coincides with that obtained from Haar-random unitary processing; (iii) this operating point is associated with local, not global, entanglement and with Lieb–Robinson velocity vLR = 1, so the regime is compatible with efficient tensor-network simulation and shallow random circuits. The second thread (Chapters 7–8) develops a modified ResNet-18 for classifying single-vertex vs pile-up events in simulated IWCD water-Cherenkov images, reporting AUC close to unity, plus an exploratory QELM application to the same neutrino data. The thesis explicitly disclaims evidence for gene

Significance. If the central XX-QELM claim is correct, the thesis makes a useful and honest contribution: the useful operating point of these QELMs is reached at a local-entanglement timescale, before volume-law entanglement develops, so the reported accuracies are not evidence of a generic quantum advantage. The use of Haar-random unitary benchmarks, shallow random circuits, K-means/ARI geometry probes, and the explicit negative framing are strengths, as is the numerical consistency between the figures and the described claims. The neutrino part is a plausible proof-of-concept for IWCD event classification. However, the dynamical-origin and classical-simulability conclusions are not as firmly established as the text suggests; the required controls are missing, and the 'efficient tensor-network simulation' statement significantly overreaches unless the readout structure is also constrained. The significance is therefore conditional on strengthening these two points.

major comments (4)
  1. [Sec. 6.3.1, 6.4.2, App. A.1] The identification of t*≈1 with the build-up of local entanglement and with the Lieb–Robinson velocity vLR = 1 is correlational. The XX Hamiltonian in Eq. (6.1) has coupling amplitude 1/2, so t*≈1 is also simply the inverse energy scale of the model. No scaling control is reported: if H is replaced by λH, does t* scale as 1/λ? If the dense-angle encoding ranges (Sec. 4.4) are narrowed or the latent dimension d/N is changed, does t* remain fixed? Without such tests, the load-bearing claim that the transition is governed by local dynamics rather than by Hamiltonian bandwidth, encoding statistics, or readout saturation is not established.
  2. [Sec. 6.6.1 and Ch. 1, pp. 4–5] The statement that the QELM operating regime 'remains compatible with efficient tensor-network simulation' overstates what is shown. The MPS representation of |ψ_t(z)> with bounded bond dimension at fixed t does not make the full QELM classically simulable, because the readout (Sec. 4.7) is a dense linear function over all 2^N outcome probabilities. Evaluating the logit requires computing sum_s W_{c,s} |<s|ψ_t(z)>|^2 for arbitrary trained weights W; this is an exponential sum unless the weights are constrained (sparse, low-rank, MPO, or otherwise structured), and no such constraint is imposed. The thesis should either restrict the simulability claim to the quantum feature map or prove that the trained readout can be contracted efficiently.
  3. [Sec. 6.3.2, Figs. 6.1–6.3] The equality of the XX plateau with Haar-random unitary processing is interpreted as showing that the specific structure of the XX Hamiltonian is irrelevant once the plateau is reached. An alternative explanation is that the plateau is a readout-capacity bound: for the fixed dense-angle encoded states and a linear readout on the 2^N probability simplex, any sufficiently mixing unitary could yield the same limiting accuracy A*(N). The paper does not include a classical random-feature baseline with the same 2^N-dimensional feature space (e.g., random Fourier features, a fixed random neural net, or a random linear map applied to the latent vector). Adding such a baseline would separate quantum-dynamics saturation from generic high-dimensional feature saturation.
  4. [Sec. 5.3.3, Fig. 5.8] The first QELM study claims that the interacting quantum reservoir 'systematically improves' accuracy over classical processing, but the only classical comparisons are one-layer neural networks on raw pixels (784+ONN) or on the 2N-dimensional autoencoder latents (AE1+ONN). These baselines do not have the nonlinear capacity or feature dimensionality of the 2^N-dimensional probability readout, so they are not representative classical image classifiers. A fair comparison would use a classical nonlinear feature map of comparable dimensionality, such as random kitchen sinks, kernel ridge regression, or a CNN on the same latent inputs. This does not affect the negative quantum-advantage conclusion of Chapter 6, but it weakens the positive 'quantum reservoir enriches representation' conclusion of Chapter 5.
minor comments (6)
  1. [Table 4.1] The amplitude-encoding row lists '2N' features per qubit; amplitude encoding embeds 2^N features in N qubits, so the entry should be 2^N (or '2^N/N' per qubit).
  2. [Sec. 2.7.4] A verbatim excerpt from Ref. [50] ('are comparably good or better than the constructed solution...') appears unquoted after Eq. (2.21). If intentional, add quotation marks and a citation; if it is a formatting artifact, remove it.
  3. [Figs. 6.1–6.3] The legends use 'Random Matrix' while the text says 'Haar-random unitary'; please unify the terminology.
  4. [Sec. 4.3.2 and figure captions] Typographical issues: 'powerfullinear' missing a space in Sec. 4.3.2, and several captions read 'T esting' instead of 'Testing'.
  5. [Eq. (6.5)] The scaling law is written as S_N(t) ≃ N f(t/N), but the preceding notation is S_{N/2}(t) for the half-chain entropy. Define S_N explicitly or write the equation in terms of the half-chain entropy to avoid confusion.
  6. [General] The manuscript contains no statement about code or data availability. For a computational thesis with many numerical experiments, a repository link or a data-availability statement would improve reproducibility.

Circularity Check

0 steps flagged

No significant circularity found; central results rest on independent numerical benchmarks and analytic bounds.

full rationale

The thesis's QELM claims are supported by explicit simulations and comparisons that are not fitted to the target result: the accuracy-versus-time curves, the Haar-random unitary benchmark (Sec. 6.3.2), the one-qubit and half-chain entropy dynamics, the Lieb–Robinson velocity computed for the XX chain (Appendix A.1), and the shallow random-circuit comparison (Sec. 6.6.2) are each computed from the model or from independent baselines. The identification of the accuracy transition time t* with local entanglement is an empirical consistency argument — both the accuracy rise and the single-qubit entropy saturation occur near t ≈ 1 — rather than a reduction by construction. The paper cites the author's prior works [19,20], but those results are re-presented in the thesis with their numerical details, so the self-citations are structural, not load-bearing. No prediction is produced by renaming a fitted parameter, and no central claim is defined in terms of its own conclusion. The neutrino classification results are similarly self-contained, being benchmarked on simulated detector data with no circular dependence on the QELM analysis.

Axiom & Free-Parameter Ledger

5 free parameters · 7 axioms · 0 invented entities

The QELM analysis introduces no new entities; its inputs are standard quantum formalism plus prior-literature Hamiltonians. The main ledger burden is in free parameters: hand-chosen evolution times, encoding rescaling ranges, AE bottleneck size, and the 10-sample Haar benchmark. The neutrino half rests on the stated-but-untested premise that simulation stands in for hardware. No invented entities (no new particles, forces, or conserved quantities) appear.

free parameters (5)
  • Evolution time (Delta t / t scan) = Delta t=20 (H2-H6), Delta t=50T (H1); t in [1e-3, 1e2] (Ch. 6)
    Hand-chosen hyperparameter and the thesis's main control variable. The transition scale t* about 1 is observed, not fitted, but selecting the post-transition plateau as the operating point is a choice.
  • Autoencoder latent dimension d = 20 (AE1/AE2 architecture dumps, Figs 5.2-5.3)
    Sets how many classical features enter the quantum register. Text also states d=2N per N (up to 24), without explaining the mapping between the fixed 20-dim dump and N>=11.
  • Hamiltonian couplings and fields H1-H6 = e.g. J0=0.06, alpha=1.51, B1=3.05 (H1); J3=-1, B3x=0.7, B3z=1.5 (H3); J4=2, J4z=0.54 (H4); B6, W in 0.02-60 (H6)
    Imported from Refs [21,31,67], not fitted to accuracy; chosen to realize specific dynamical regimes (integrable, chaotic, MBL). The thesis's conclusion that Hamiltonian details do not matter is conditional on these prior parameter choices.
  • Encoding rescaling ranges = [0,pi] polar, [0,2pi] azimuthal (dense-angle); [0,1] populations (uniform Bloch)
    Hand-chosen feature-to-Bloch-sphere maps; the comparison among encoding strategies depends on these ranges.
  • Haar-random benchmark samples = 10 unitaries per (N, dataset)
    Averaging over 10 random unitaries; the shaded band is the standard deviation over these 10 only.
axioms (7)
  • standard math Lieb-Robinson bound with vLR=1 for the nearest-neighbour XX chain
    Appendix A.1; used to conclude the accuracy transition at t* about 1 is local (Sec. 6.4.2).
  • domain assumption Computational-basis measurement is sufficiently misaligned with the XX eigenbasis to generate time-dependent, expressive features
    Sec. 4.6; the entire time-dependence analysis (Eq. 4.33) requires this misalignment, asserted because XX eigenstates are delocalized over the chain.
  • domain assumption Reconstruction-trained autoencoder latent space preserves class-discriminative information
    Sec. 4.3.2 and 5.3.1; using reconstruction loss as a proxy for task-relevant information is assumed; downstream accuracy is the only empirical check.
  • domain assumption Simulated IWCD detector images (event generation plus detector simulation, Sec. 8.3) faithfully represent real detector response
    The near-unity ResNet AUC (Sec. 8.4.6) is defined entirely on simulation; transfer to hardware is not demonstrated.
  • domain assumption Haar-random unitaries are the appropriate maximal-genericity benchmark for the encoding/readout pipeline
    Sec. 6.3.2; the plateau-coincidence claim gives this benchmark its meaning.
  • standard math Constant-depth Trotterized local evolution implies efficient tensor-network simulability at the relevant accuracy
    Sec. 6.6.1; standard MPS/rTEBD reasoning connecting t* about 1 to constant circuit depth.
  • standard math Born rule / state-vector formalism of quantum mechanics
    Foundation of the QELM measurement model (Eqs. 4.28 and 6.3).

pith-pipeline@v1.3.0-alltime-deepseek · 48589 in / 24924 out tokens · 236490 ms · 2026-08-02T04:01:12.195876+00:00 · methodology

0 comments
read the original abstract

This thesis investigates the application of machine-learning methods in the context of quantum computing and neutrino physics, with particular emphasis on the construction of effective representations for complex, high-dimensional data. The first part of the work is devoted to Quantum Extreme Learning Machines (QELMs), a hybrid quantum--classical framework in which classical data are encoded into quantum states and processed through fixed quantum dynamics, while learning is performed by a classical readout layer. Within this framework, we analyze the role of encoding strategies, feature-reduction methods, Hamiltonian structure, and measurement, with particular focus on the relationship between quantum dynamics, expressivity, entanglement, and classical simulability. The second part of the thesis concerns the application of deep learning to the analysis of images produced by water Cherenkov detectors in neutrino physics. Convolutional architectures, including residual networks, are developed for the classification of complex events in realistic simulated datasets, showing that such models can effectively extract relevant information from detector data. Taken together, these results highlight the potential of machine learning, in both its classical and quantum forms, as a powerful framework for the analysis of complex data in fundamental physics, while also outlining relevant challenges and directions for future research.

Figures

Figures reproduced from arXiv: 2607.13699 by Annalisa De Lorenzis.

Figure 2.1
Figure 2.1. Figure 2.1: Some non-linear activation functions. Both saturating (top row) and [PITH_FULL_IMAGE:figures/full_fig_p022_2_1.png] view at source ↗
Figure 2.2
Figure 2.2. Figure 2.2: A representation of a feed-forward neural network with two hidden [PITH_FULL_IMAGE:figures/full_fig_p023_2_2.png] view at source ↗
Figure 2.3
Figure 2.3. Figure 2.3: Convolution of a multi-channel image with [PITH_FULL_IMAGE:figures/full_fig_p031_2_3.png] view at source ↗
Figure 2.4
Figure 2.4. Figure 2.4: Example of a 2×2 max-pooling operation. Each 2×2 region of the input feature map is replaced by its maximum value, producing a downsampled output. In decoder-like parts of convolutional architectures, such as the decoder of a convolutional autoencoder, one often needs to increase the spatial resolution of the feature maps, effectively inverting the action of pooling or strided convolutions. This is achie… view at source ↗
Figure 2.5
Figure 2.5. Figure 2.5: Architecture of a typical CNN. Neurons in the convolutional layers [PITH_FULL_IMAGE:figures/full_fig_p033_2_5.png] view at source ↗
Figure 2
Figure 2. Figure 2: Residual learning: a building block. are comparably good or better than the constructed solution (or unable to do so in feasible time). In this paper, we address the degradation problem by introducing a deep residual learning framework. In￾stead of hoping each few stacked layers directly fit a desired underlying mapping, we explicitly let these lay￾ers fit a residual mapping. Formally, denoting the desired… view at source ↗
Figure 3.1
Figure 3.1. Figure 3.1: Geometric representation of a single-qubit state and schematic illustra [PITH_FULL_IMAGE:figures/full_fig_p040_3_1.png] view at source ↗
Figure 3.2
Figure 3.2. Figure 3.2: Schematic comparison between gate-based digital quantum computation [PITH_FULL_IMAGE:figures/full_fig_p042_3_2.png] view at source ↗
Figure 3.3
Figure 3.3. Figure 3.3: Schematic representation of the quantum–classical processing pipeline [PITH_FULL_IMAGE:figures/full_fig_p045_3_3.png] view at source ↗
Figure 4.1
Figure 4.1. Figure 4.1: A schematic representation of the QELM. The workflow is as follows: [PITH_FULL_IMAGE:figures/full_fig_p050_4_1.png] view at source ↗
Figure 4.2
Figure 4.2. Figure 4.2: Correlated 2D data and the corresponding principal components. Drop [PITH_FULL_IMAGE:figures/full_fig_p052_4_2.png] view at source ↗
Figure 4.3
Figure 4.3. Figure 4.3: Visualization of PCA as a rotation and projection in two dimensions. [PITH_FULL_IMAGE:figures/full_fig_p053_4_3.png] view at source ↗
Figure 4.4
Figure 4.4. Figure 4.4: Schematic representation of an autoencoder architecture. The input [PITH_FULL_IMAGE:figures/full_fig_p056_4_4.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Dynamics of observables [PITH_FULL_IMAGE:figures/full_fig_p067_2.png] view at source ↗
Figure 5.1
Figure 5.1. Figure 5.1: Sample images from three widely used benchmark datasets in this study. [PITH_FULL_IMAGE:figures/full_fig_p074_5_1.png] view at source ↗
Figure 5.2
Figure 5.2. Figure 5.2: Architecture of the first autoencoder explored in the present analysis. [PITH_FULL_IMAGE:figures/full_fig_p077_5_2.png] view at source ↗
Figure 5.3
Figure 5.3. Figure 5.3: Architecture of the second autoencoder explored in the present analy [PITH_FULL_IMAGE:figures/full_fig_p078_5_3.png] view at source ↗
Figure 5.4
Figure 5.4. Figure 5.4: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p079_5_4.png] view at source ↗
Figure 5.5
Figure 5.5. Figure 5.5: Training (solid line) and testing (dashed line) accuracy as a function [PITH_FULL_IMAGE:figures/full_fig_p080_5_5.png] view at source ↗
Figure 5.6
Figure 5.6. Figure 5.6: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p082_5_6.png] view at source ↗
Figure 5.7
Figure 5.7. Figure 5.7: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p082_5_7.png] view at source ↗
Figure 5.8
Figure 5.8. Figure 5.8: Training (left panel) and testing (right panel) accuracy as a function [PITH_FULL_IMAGE:figures/full_fig_p083_5_8.png] view at source ↗
Figure 5.9
Figure 5.9. Figure 5.9: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p085_5_9.png] view at source ↗
Figure 6.1
Figure 6.1. Figure 6.1: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p093_6_1.png] view at source ↗
Figure 6.2
Figure 6.2. Figure 6.2: Training (left panel) and testing (right panel) accuracy as a function [PITH_FULL_IMAGE:figures/full_fig_p094_6_2.png] view at source ↗
Figure 6.3
Figure 6.3. Figure 6.3: Training (left panel) and testing (right panel) accuracy as a function [PITH_FULL_IMAGE:figures/full_fig_p094_6_3.png] view at source ↗
Figure 6.4
Figure 6.4. Figure 6.4: a) Von Neumann entropy as function of the evolution time of half [PITH_FULL_IMAGE:figures/full_fig_p098_6_4.png] view at source ↗
Figure 6.5
Figure 6.5. Figure 6.5: Accuracy (in magenta color) and Inertia (in light blue color) as a function [PITH_FULL_IMAGE:figures/full_fig_p102_6_5.png] view at source ↗
Figure 6.6
Figure 6.6. Figure 6.6: a) Accuracy (in magenta color) and Inertia (in light blue color) as a func [PITH_FULL_IMAGE:figures/full_fig_p104_6_6.png] view at source ↗
Figure 6.7
Figure 6.7. Figure 6.7: A schematic representation of an algorithm with 5 qubits, in which [PITH_FULL_IMAGE:figures/full_fig_p106_6_7.png] view at source ↗
Figure 6.8
Figure 6.8. Figure 6.8: Training (left panel) and testing (right panel) accuracy as a function of [PITH_FULL_IMAGE:figures/full_fig_p108_6_8.png] view at source ↗
Figure 6.9
Figure 6.9. Figure 6.9: Training (left panel) and testing (right panel) accuracy as a function [PITH_FULL_IMAGE:figures/full_fig_p108_6_9.png] view at source ↗
Figure 7.1
Figure 7.1. Figure 7.1: Schematic illustration of a Cherenkov light cone produced by a charged [PITH_FULL_IMAGE:figures/full_fig_p119_7_1.png] view at source ↗
Figure 7.2
Figure 7.2. Figure 7.2: Schematic comparison of the Kamiokande, Super-Kamiokande, and [PITH_FULL_IMAGE:figures/full_fig_p122_7_2.png] view at source ↗
Figure 7
Figure 7. Figure 7: mPMTs in unwrapped ta See Poster by N. Deshmuk: Mechanical Design of MultiPMTs for IWC Figure 7: PMTid t See Poster by N. Deshmuk: Mechanical Design of Multi-PMTs for IWCD Resnet-18 536 mPMT ee Poster by NDeshmuk: Mechanical Design of Multi-PMTs for IWCD 536 mPMT [PITH_FULL_IMAGE:figures/full_fig_p124_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: mPMTs in unwrapped ta See Poster by N. Deshmuk: Mechanical Design of MultiPMTs for IWC Figure 7: PMTid t See Poster by N. Deshmuk: Mechanical Design of Multi-PMTs for IWCD Resnet-18 536 mPMT ee Poster by NDeshmuk: Mechanical Design of Multi-PMTs for IWCD [PITH_FULL_IMAGE:figures/full_fig_p125_7.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between fiTQun and ML performance in 𝑄/electron separation. Motivated by the performance of the ML framework in distinguishing particle gun events (𝑂→, 𝑃→, 𝑄, and 𝐿0), we applied it to select 𝑀𝐿 events from a simulated beam profile. This IWCD simulated dataset contains single-ring CCQE (𝑀𝐿𝑁𝑁0𝐿, 𝑀𝑀𝑁𝑁), NC (NC 𝑄, NC𝐿0), and other events such as more complex CC events with pions and/or multiple oth… view at source ↗
Figure 8.3
Figure 8.3. Figure 8.3: Examples of more realistic event topologies illustrating the challenge [PITH_FULL_IMAGE:figures/full_fig_p131_8_3.png] view at source ↗
Figure 1
Figure 1. Figure 1: Configuration of the cylindrical detector onto a 2D image for CNN. The layout illustration on the left demonstrates how the cylinder is unwrapped onto a 2D surface. Shown on the right, the re￾sulting image is divided into sections outlined by dashed lines, then duplicated and reconfigured into a double cover of the detector surface, with circular boundary conditions indicated by orange arrows. The use of C… view at source ↗
Figure 8.5
Figure 8.5. Figure 8.5: Overall ResNet-18–based architecture used in this thesis. The stem is [PITH_FULL_IMAGE:figures/full_fig_p138_8_5.png] view at source ↗
Figure 8.6
Figure 8.6. Figure 8.6: Input stem: Conv 1 × 1 followed by BN and a ReLU. The output is passed to the first residual stage L1. 8.4.1 Network Stem The network stem performs the initial transformation of the input tensor into a fea￾ture representation. Unlike the canonical ImageNet ResNet-18 architecture, which uses a 7 × 7 convolution with stride s = 2 followed by max pooling, the present implementation employs a simplified stem… view at source ↗
Figure 8.7
Figure 8.7. Figure 8.7: Residual stages and downsampling schedule. Downsampling by a factor [PITH_FULL_IMAGE:figures/full_fig_p139_8_7.png] view at source ↗
Figure 8.8
Figure 8.8. Figure 8.8: Residual block used in the network: two 3×3 convolutions with BN and ReLU, summed with an identity shortcut. When stride or channel dimensionality changes, the shortcut uses a 1 × 1 projection (dashed). 4. A final ReLU activation. The two 3×3 convolutions in the residual block are characterized by a padding of one pixel and no bias term. We remark that this operator preserves spatial resolution for unit … view at source ↗
Figure 8.9
Figure 8.9. Figure 8.9: Output head: adaptive global average pooling followed by a fully con [PITH_FULL_IMAGE:figures/full_fig_p141_8_9.png] view at source ↗
Figure 8.10
Figure 8.10. Figure 8.10: Receiver operating characteristic (ROC) curves for the single-vertex [PITH_FULL_IMAGE:figures/full_fig_p144_8_10.png] view at source ↗
Figure 8.12
Figure 8.12. Figure 8.12: Confusion matrices for the single-vertex versus pile-up classification [PITH_FULL_IMAGE:figures/full_fig_p145_8_12.png] view at source ↗
Figure 8.13
Figure 8.13. Figure 8.13: Classification accuracy for the 19Q+19T configuration as a function of event-level quantities, shown for multi-vertex events with exactly two tracks. Results are compared across the No-Physics, GENIE, and NEUT datasets [PITH_FULL_IMAGE:figures/full_fig_p146_8_13.png] view at source ↗
Figure 8.14
Figure 8.14. Figure 8.14: Classification accuracy for the 19Q+19T configuration as a function of geometry-related quantities. The dependence on distances to the detector wall highlights boundary effects that can impact the reconstruction of Cherenkov light patterns [PITH_FULL_IMAGE:figures/full_fig_p147_8_14.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

153 extracted references · 8 linked inside Pith

  1. [1]

    Bishop.Pattern Recognition and Machine Learning

    Christopher M. Bishop.Pattern Recognition and Machine Learning. Springer, New York, 2006

    2006

  2. [2]

    MIT Press, Cambridge, MA, 2016

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville.Deep Learning. MIT Press, Cambridge, MA, 2016

    2016

  3. [3]

    O’Reilly Media, 3 edition, 2022

    Aurélien Géron.Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow. O’Reilly Media, 3 edition, 2022

    2022

  4. [4]

    Richard P. Feynman. Simulating physics with computers.Int. J. Theor. Phys., 21:467–488, 1982

    1982

  5. [5]

    Nielsen and Isaac L

    Michael A. Nielsen and Isaac L. Chuang.Quantum Computation and Quan- tum Information: 10th Anniversary Edition. Cambridge University Press, Cambridge, 2010

    2010

  6. [6]

    Quantum algorithms: an overview.npj Quantum Inf., 2:15023, 2016

    Ashley Montanaro. Quantum algorithms: an overview.npj Quantum Inf., 2:15023, 2016

    2016

  7. [7]

    Quantum computing in the NISQ era and beyond.Quantum, 2:79, 2018

    John Preskill. Quantum computing in the NISQ era and beyond.Quantum, 2:79, 2018

    2018

  8. [8]

    Bardin, Rami Barends, Rupak Biswas, Sergio Boixo, Fernando G

    Frank Arute, Kunal Arya, Ryan Babbush, Dave Bacon, Joseph C. Bardin, Rami Barends, Rupak Biswas, Sergio Boixo, Fernando G. S. L. Brandao, David A. Buell, et al. Quantum supremacy using a programmable super- conducting processor.Nature, 574:505–510, 2019

    2019

  9. [9]

    Quantum advantage in learning from experiments.Science, 376:1182–1186, 2022

    Hsin-Yuan Huang, Michael Broughton, Jordan Cotler, Sitan Chen, Jerry Li, Masoud Mohseni, Hartmut Neven, Ryan Babbush, Richard Kueng, John BIBLIOGRAPHY 160 Preskill, et al. Quantum advantage in learning from experiments.Science, 376:1182–1186, 2022

    2022

  10. [10]

    Quantum machine learning.Nature, 549:195–202, 2017

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning.Nature, 549:195–202, 2017

    2017

  11. [11]

    Quantum Science and Technology

    Maria Schuld and Francesco Petruccione.Supervised Learning with Quantum Computers. Quantum Science and Technology. Springer, Cham, 2018

    2018

  12. [12]

    Quantum machine learning in feature Hilbert spaces.Phys

    Maria Schuld and Nathan Killoran. Quantum machine learning in feature Hilbert spaces.Phys. Rev. Lett., 122:040504, 2019

    2019

  13. [13]

    Springer, Cham, 2 edition, 2021

    Maria Schuld and Francesco Petruccione.Machine Learning with Quantum Computers. Springer, Cham, 2 edition, 2021

    2021

  14. [14]

    Iris Cong, Soonwon Choi, and Mikhail D. Lukin. Quantum convolutional neural networks.Nat. Phys., 15:1273–1278, 2019

    2019

  15. [15]

    Quantum capsule networks.Quantum Sci

    Zidu Liu, Pei-Xin Shen, Weikang Li, Lu-Ming Duan, and Dong-Ling Deng. Quantum capsule networks.Quantum Sci. Technol., 8:015016, 2023

    2023

  16. [16]

    Quantum algorithms for supervised and unsupervised machine learning, 2013

    Seth Lloyd, Masoud Mohseni, and Patrick Rebentrost. Quantum algorithms for supervised and unsupervised machine learning, 2013. arXiv:1307.0411 [quant-ph]

    Pith/arXiv arXiv 2013

  17. [17]

    The quest for a quantum neural network.Quantum Inf

    Maria Schuld, Ilya Sinayskiy, and Francesco Petruccione. The quest for a quantum neural network.Quantum Inf. Process., 13:2567–2586, 2014

    2014

  18. [18]

    Córcoles, Kristan Temme, Aram W

    Vojtěch Havlíček, Antonio D. Córcoles, Kristan Temme, Aram W. Harrow, Abhinav Kandala, Jerry M. Chow, and Jay M. Gambetta. Supervised learning with quantum-enhanced feature spaces.Nature, 567:209–212, 2019

    2019

  19. [19]

    Estarellas, Nicola Lo Gullo, Thorsten Lux, Francesco Plastina, Arnau Riera, and Jacopo Set- tino

    Annalisa De Lorenzis, Maria Pilar Casado, Marta P. Estarellas, Nicola Lo Gullo, Thorsten Lux, Francesco Plastina, Arnau Riera, and Jacopo Set- tino. Harnessing quantum extreme learning machines for image classification. Phys. Rev. Applied, 23:044024, 2025. BIBLIOGRAPHY 161

    2025

  20. [20]

    Entanglement and classical simulability in quantum extreme learning machines.Phys

    Annalisa De Lorenzis, Maria Pilar Casado, Nicola Lo Gullo, Thorsten Lux, Francesco Plastina, and Arnau Riera. Entanglement and classical simulability in quantum extreme learning machines.Phys. Rev. Applied, 25:054052, 2026

    2026

  21. [21]

    Estarellas, William J

    Akitada Sakurai, Marta P. Estarellas, William J. Munro, and Kae Nemoto. Quantum extreme reservoir computation utilizing scale-free networks.Phys. Rev. Applied, 17:064044, 2022

    2022

  22. [22]

    Munro, and Kae Nemoto

    Aoi Hayashi, Akitada Sakurai, Shin Nishio, William J. Munro, and Kae Nemoto. Impact of the form of weighted networks on the quantum extreme reservoir computation.Phys. Rev. A, 108:042609, 2023

    2023

  23. [23]

    Munro, and Kae Nemoto

    Aoi Hayashi, Akitada Sakurai, William J. Munro, and Kae Nemoto. Effective quantum feature maps in quantum extreme reservoir computation from the XY model.Phys. Rev. A, 111:022431, 2025

    2025

  24. [24]

    Munro, and Kae Nemoto

    Akitada Sakurai, Aoi Hayashi, William J. Munro, and Kae Nemoto. Simple Hamiltonian dynamics as a powerful resource for image classification.Phys. Rev. A, 111:052432, 2025

    2025

  25. [25]

    Experimental property reconstruction in a photonic quantum extreme learning machine.Phys

    Alessia Suprano, Danilo Zia, Luca Innocenti, Salvatore Lorenzo, Valeria Ci- mini, Taira Giordani, Ivan Palmisano, Emanuele Polino, Nicolò Spagnolo, Fabio Sciarrino, et al. Experimental property reconstruction in a photonic quantum extreme learning machine.Phys. Rev. Lett., 132:160802, 2024

    2024

  26. [26]

    Massimo Palma

    Marco Vetrano, Gabriele Lo Monaco, Luca Innocenti, Salvatore Lorenzo, and G. Massimo Palma. State estimation with quantum extreme learning machines beyond the scrambling time.npj Quantum Inf., 11:20, 2025

    2025

  27. [27]

    Massimo Palma

    Luca Innocenti, Salvatore Lorenzo, Ivan Palmisano, Alessandro Ferraro, Mauro Paternostro, and G. Massimo Palma. Potential and limitations of quantum extreme learning machines.Commun. Phys., 6:118, 2023

    2023

  28. [28]

    Khan, and Hakan E

    Gerasimos Angelatos, Saeed A. Khan, and Hakan E. Türeci. Reservoir com- puting approach to quantum state measurement.Phys. Rev. X, 11:041062, 2021. BIBLIOGRAPHY 162

    2021

  29. [29]

    Soriano, and Roberta Zambrini

    Pere Mujal, Johannes Nokkala, Rodrigo Martínez-Peña, Gian Luca Giorgi, Miguel C. Soriano, and Roberta Zambrini. Analytical evidence of nonlinear- ity in qubits and continuous-variable quantum reservoir computing.J. Phys. Complex., 2:045008, 2021

    2021

  30. [30]

    Carlo, and Florentino Borondo

    Laia Domingo, Gabriel G. Carlo, and Florentino Borondo. Optimal quantum reservoir computing for the noisy intermediate-scale quantum era.Phys. Rev. E, 106:L043301, 2022

    2022

  31. [31]

    Sori- ano, and Roberta Zambrini

    Rodrigo Martínez-Peña, Gian Luca Giorgi, Johannes Nokkala, Miguel C. Sori- ano, and Roberta Zambrini. Dynamical phase transitions in quantum reservoir computing.Phys. Rev. Lett., 127:100502, 2021

    2021

  32. [32]

    Harnessing disordered-ensemble quantum dynamics for machine learning.Phys

    Keisuke Fujii and Kohei Nakajima. Harnessing disordered-ensemble quantum dynamics for machine learning.Phys. Rev. Applied, 8:024030, 2017

    2017

  33. [33]

    Nurdin, and Naoki Yamamoto

    Jiayin Chen, Hendra I. Nurdin, and Naoki Yamamoto. Temporal information processing on noisy quantum computers.Phys. Rev. Applied, 14:024065, 2020

    2020

  34. [34]

    Extreme learning machine: theory and applications.Neurocomputing, 70:489–501, 2006

    Guang-Bin Huang, Qin-Yu Zhu, and Chee-Kheong Siew. Extreme learning machine: theory and applications.Neurocomputing, 70:489–501, 2006

    2006

  35. [35]

    Extreme learning ma- chines: a survey.Int

    Guang-Bin Huang, Dian-Hui Wang, and Yuan Lan. Extreme learning ma- chines: a survey.Int. J. Mach. Learn. Cybern., 2:107–122, 2011

    2011

  36. [36]

    A review on extreme learning machine.Multimed

    Jian Wang, Siyuan Lu, Shui-Hua Wang, and Yu-Dong Zhang. A review on extreme learning machine.Multimed. Tools Appl., 81:41611–41660, 2022

    2022

  37. [37]

    Extreme learning machine versus classical feedforward network.Neural Comput

    Urszula Markowska-Kaczmar and Michał Kosturek. Extreme learning machine versus classical feedforward network.Neural Comput. Appl., 33:15121–15144, 2021

    2021

  38. [38]

    Extreme learning machine: algorithm, theory and applications.Artif

    Shifei Ding, Han Zhao, Yanan Zhang, Xinzheng Xu, and Ru Nie. Extreme learning machine: algorithm, theory and applications.Artif. Intell. Rev., 44:103–115, 2015. BIBLIOGRAPHY 163

    2015

  39. [39]

    Trends in extreme learning machines: a review.Neural Networks, 61:32–48, 2015

    Gao Huang, Guang-Bin Huang, Shiji Song, and Keyou You. Trends in extreme learning machines: a review.Neural Networks, 61:32–48, 2015

    2015

  40. [40]

    Avariational eigenvalue solver on a photonic quantum processor.Nat

    Alberto Peruzzo, Jarrod McClean, Peter Shadbolt, Man-Hong Yung, Xiao-Qi Zhou, PeterJ.Love, AlánAspuru-Guzik, andJeremyL.O’Brien. Avariational eigenvalue solver on a photonic quantum processor.Nat. Commun., 5:4213, 2014

    2014

  41. [41]

    Aquantumapproximate optimization algorithm, 2014

    EdwardFarhi, JeffreyGoldstone, andSamGutmann. Aquantumapproximate optimization algorithm, 2014. arXiv:1411.4028 [quant-ph]

    Pith/arXiv arXiv 2014

  42. [42]

    Benjamin, Sug- uru Endo, Keisuke Fujii, Jarrod R

    Marco Cerezo, Andrew Arrasmith, Ryan Babbush, Simon C. Benjamin, Sug- uru Endo, Keisuke Fujii, Jarrod R. McClean, Kosuke Mitarai, Xiao Yuan, Lukasz Cincio, et al. Variational quantum algorithms.Nat. Rev. Phys., 3:625– 644, 2021

    2021

  43. [43]

    Yann LeCun, Corinna Cortes, and Christopher J. C. Burges. MNIST hand- written digit database. AT&T Labs [Online]. Available:http://yann.lecun. com/exdb/mnist, 2010

    2010

  44. [44]

    Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms, 2017

    Han Xiao, Kashif Rasul, and Roland Vollgraf. Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms, 2017. arXiv:1708.07747 [cs.LG]

    Pith/arXiv arXiv 2017

  45. [45]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Tech- nical report, Department of Computer Science, University of Toronto, 2009. Available:https://www.cs.toronto.edu/~kriz/cifar.html

    2009

  46. [46]

    Abe, et al

    Hyper-Kamiokande Collaboration, K. Abe, et al. Hyper-Kamiokande Design Report, 2018. arXiv:1805.04163 [physics.ins-det]

    Pith/arXiv arXiv 2018

  47. [47]

    Abe, et al

    Hyper-Kamiokande Collaboration, K. Abe, et al. Physics potential of a long- baseline neutrino oscillation experiment using a J-PARC neutrino beam and Hyper-Kamiokande.Prog. Theor. Exp. Phys., 2015:053C02, 2015. BIBLIOGRAPHY 164

    2015

  48. [48]

    Abe, et al

    Hyper-Kamiokande Collaboration, K. Abe, et al. Sensitivity of the Hyper- Kamiokande experiment to neutrino oscillation parameters using accelerator neutrinos.Eur. Phys. J. C, 86:170, 2026

    2026

  49. [49]

    Abe, et al

    Hyper-Kamiokande Collaboration, K. Abe, et al. The Hyper-Kamiokande experiment: input to the update of the European Strategy for Particle Physics,

  50. [50]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProc. IEEE Conf. Comput. Vis. Pattern Recognit., pages 770–778, 2016

    2016

  51. [51]

    WatChMaL: Water Cherenkov machine learning

    WatChMaL Collaboration. WatChMaL: Water Cherenkov machine learning. GitHub repository. Available:https://github.com/WatChMaL/WatChMaL, 2020

    2020

  52. [52]

    Arthur L. Samuel. Some studies in machine learning using the game of check- ers.IBM J. Res. Dev., 3:210–229, 1959

    1959

  53. [53]

    Wiley, New York, 1949

    DonaldO.Hebb.The Organization of Behavior: A Neuropsychological Theory. Wiley, New York, 1949

    1949

  54. [54]

    McCulloch and Walter Pitts

    Warren S. McCulloch and Walter Pitts. A logical calculus of the ideas imma- nent in nervous activity.Bull. Math. Biophys., 5:115–133, 1943

    1943

  55. [55]

    Rumelhart, Geoffrey E

    David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams. Learning representations by back-propagating errors.Nature, 323:533–536, 1986

    1986

  56. [56]

    Support-vector networks.Mach

    Corinna Cortes and Vladimir Vapnik. Support-vector networks.Mach. Learn., 20:273–297, 1995

    1995

  57. [57]

    Deep learning.Nature, 521:436–444, 2015

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning.Nature, 521:436–444, 2015

    2015

  58. [58]

    Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda BIBLIOGRAPHY 165 Panneershelvam, Marc Lanctot, et al

    David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda BIBLIOGRAPHY 165 Panneershelvam, Marc Lanctot, et al. Mastering the game of Go with deep neural networks and tree search.Nature, 529:484–489, 2016

    2016

  59. [59]

    Pankaj Mehta, Marin Bukov, Ching-Hao Wang, Alexandre G. R. Day, Clint Richardson, CharlesK.Fisher, andDavidJ.Schwab. Ahigh-bias, low-variance introduction to machine learning for physicists.Phys. Rep., 810:1–124, 2019

    2019

  60. [60]

    Approximation by superpositions of a sigmoidal function

    George Cybenko. Approximation by superpositions of a sigmoidal function. Math. Control Signal Systems, 2:303–314, 1989

    1989

  61. [61]

    Multilayer feed- forward networks are universal approximators.Neural Networks, 2:359–366, 1989

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feed- forward networks are universal approximators.Neural Networks, 2:359–366, 1989

    1989

  62. [62]

    Gradient- based learning applied to document recognition.Proc

    Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient- based learning applied to document recognition.Proc. IEEE, 86:2278–2324, 1998

    1998

  63. [63]

    Cam- bridge University Press, Cambridge, 3 edition, 2020

    Jun John Sakurai and Jim Napolitano.Modern Quantum Mechanics. Cam- bridge University Press, Cambridge, 3 edition, 2020

    2020

  64. [64]

    echo state

    Herbert Jaeger. The “echo state” approach to analysing and training recurrent neural networks. GMD Report 148, German National Research Center for Information Technology, 2001

    2001

  65. [65]

    Real-time com- puting without stable states: a new framework for neural computation based on perturbations.Neural Comput., 14:2531–2560, 2002

    Wolfgang Maass, Thomas Natschläger, and Henry Markram. Real-time com- puting without stable states: a new framework for neural computation based on perturbations.Neural Comput., 14:2531–2560, 2002

    2002

  66. [66]

    Soriano, and Roberta Zambrini

    Pere Mujal, Rodrigo Martínez-Peña, Johannes Nokkala, Jorge García-Beni, Gian Luca Giorgi, Miguel C. Soriano, and Roberta Zambrini. Opportunities in quantum reservoir computing and extreme learning machines.Adv. Quantum Technol., 4:2100027, 2021

    2021

  67. [67]

    On fundamental aspects of quantum extreme learning machines.Quantum Mach

    Weijie Xiong, Giorgio Facelli, Mehrad Sahebi, Owen Agnel, Thiparat Chotibut, Supanut Thanasilp, and Zoë Holmes. On fundamental aspects of quantum extreme learning machines.Quantum Mach. Intell., 7:20, 2025. BIBLIOGRAPHY 166

    2025

  68. [68]

    Marco Cerezo, Guillaume Verdon, Hsin-Yuan Huang, Lukasz Cincio, and Patrick J. Coles. Challenges and opportunities in quantum machine learn- ing.Nat. Comput. Sci., 2:567–576, 2022

    2022

  69. [69]

    Fiderer, Hendrik Poulsen Nautrup, Jonas M

    Sofiene Jerbi, Lukas J. Fiderer, Hendrik Poulsen Nautrup, Jonas M. Kübler, Hans J. Briegel, and Vedran Dunjko. Quantum machine learning beyond kernel methods.Nat. Commun., 14:517, 2023

    2023

  70. [70]

    Hsin-Yuan Huang, Michael Broughton, Masoud Mohseni, Ryan Babbush, Ser- gio Boixo, Hartmut Neven, and Jarrod R. McClean. Power of data in quantum machine learning.Nat. Commun., 12:2631, 2021

    2021

  71. [71]

    McClean, Sergio Boixo, Vadim N

    Jarrod R. McClean, Sergio Boixo, Vadim N. Smelyanskiy, Ryan Babbush, and Hartmut Neven. Barren plateaus in quantum neural network training landscapes.Nat. Commun., 9:4812, 2018

    2018

  72. [72]

    Samson Wang, Enrico Fontana, Marco Cerezo, Kunal Sharma, Akira Sone, Lukasz Cincio, and Patrick J. Coles. Noise-induced barren plateaus in varia- tional quantum algorithms.Nat. Commun., 12:6961, 2021

    2021

  73. [73]

    Soriano, and Roberta Zambrini

    Jorge García-Beni, Iris Paparelle, Valentina Parigi, Gian Luca Giorgi, Miguel C. Soriano, and Roberta Zambrini. Quantum machine learning via continuous-variable cluster states and teleportation.EPJ Quant. Technol., 12:63, 2025

    2025

  74. [74]

    Nicoli, Paolo Stornati, Rouven Koch, Miriam Büttner, et al.Machine learning in quantum sciences

    Anna Dawid, Julian Arnold, Borja Requena, Alexander Gresch, Marcin Płodzień, Kaelan Donatella, Kim A. Nicoli, Paolo Stornati, Rouven Koch, Miriam Büttner, et al.Machine learning in quantum sciences. Cambridge University Press, 2025

    2025

  75. [75]

    La- torre

    Adrián Pérez-Salinas, Alba Cervera-Lierta, Elies Gil-Fuster, and José I. La- torre. Data re-uploading for a universal quantum classifier.Quantum, 4:226, 2020. BIBLIOGRAPHY 167

    2020

  76. [76]

    Harnessing quantum backaction for time-series processing

    Giacomo Franceschetto, Marcin Płodzień, Maciej Lewenstein, Antonio Acín, and Pere Mujal. Harnessing quantum backaction for time-series processing. Phys. Rev. X, 16:021002, 2026

    2026

  77. [77]

    An insight into extreme learning machines: random neu- rons, random features and kernels.Cogn

    Guang-Bin Huang. An insight into extreme learning machines: random neu- rons, random features and kernels.Cogn. Comput., 6:376–390, 2014

    2014

  78. [78]

    Extreme learning machine: a new learning scheme of feedforward neural networks

    Guang-Bin Huang, Qin-Yu Zhu, and Chee Kheong Siew. Extreme learning machine: a new learning scheme of feedforward neural networks. InProc. IEEE Int. Joint Conf. Neural Networks, volume 2, pages 985–990, 2004

    2004

  79. [79]

    Solanki and Anh Pham

    Payal D. Solanki and Anh Pham. Harnessing quantum dynamics for robust and scalable quantum extreme learning machines.Phys. Rev. A, 113:042443, 2026

    2026

  80. [80]

    Entanglement estimation of Werner states with a quantum extreme learning machine.Phys

    Hajar Assil, Abderrahim El Allati, and Gian Luca Giorgi. Entanglement estimation of Werner states with a quantum extreme learning machine.Phys. Rev. A, 111:022412, 2025

    2025

Showing first 80 references.