{"id":"214a613b-6b8b-4922-aa7c-a4d46438499b","arxiv_id":"2412.21082","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"On a 1,000-jet CMS subset, hybrid and fully quantum diffusion models reach FID scores of 1.81 and 2.74, close to a 1.82 classical baseline, with no error bars.","lead":"The paper trains fully quantum and hybrid diffusion models on CMS quark and gluon jet images, replacing Gaussian noise with random unitary matrices and using a variational quantum circuit for denoising. It is a test of whether quantum circuits can be dropped into generative models for physics data without losing quality, although no speedup or hardware result is shown.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"One-shot Haar-unitary forward noising preserves information, so Eq. 5 is undefined and the model is closer to an autoencoder; the central 'quantum diffusion' competitiveness claim is therefore unsupported without a derivation or ablation.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: the unitary forward process is not a valid diffusion noising process. I agree. The other issues, such as single-run FID values and absence of timing comparisons, are real but secondary; they can be patched by adding seeded repeats and cost measurements. The unitary-noising issue is more fundamental because it calls into question whether the architecture being benchmarked is a diffusion model at all. The paper's open-source code and clear architectural description are real assets, which make the proposed test directly feasible. This is not an accusation of bad faith; the model may still generate plausible jets, but the reported result cannot currently be interpreted as evidence for a quantum diffusion model. Because the flaw is reparable with an ablation and a derivation of an appropriate objective, the reader's CONDITIONAL verdict remains appropriate; my read does not move the verdict.","tokens_in":5745,"tokens_out":7134,"duration_ms":80454,"concrete_test":"Run the trained pipeline with the Haar unitary replaced by (a) a fixed classical orthogonal matrix applied once per channel and (b) standard multi-step Gaussian diffusion with T=1000, keeping the same denoiser capacity and training budget. Compare FID and loss curves. If (a) reproduces the reported FID values without any timestep schedule, the unitary transform is not functioning as diffusion; if (b) is substantially better, the one-shot unitary is not a valid generative prior. Additionally, compute the KL term in Eq. 5 for q(x_T|x_0)=delta(x_T-Ux0); if it is not finite, the paper's stated training objective cannot be the diffusion bound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.2 replaces the Gaussian forward process with a single Haar-random unitary applied to each encoded channel, justified only by a citation to cold diffusion [29]. This justification is not sufficient. A unitary is invertible and norm-preserving, so no information is destroyed and the forward conditional q(x_T|x_0) is a delta function rather than a progressively noised distribution. Consequently, the variational objective in Eq. 5 is not defined for this process: the term D_KL(q(x_T|x_0) || p(x_T)) is infinite for any continuous p(x_T). Cold diffusion [29] does not apply because it relies on a sequence of degradation operators that progressively remove information; a one-shot unitary never does. The reverse network is therefore learning to invert a fixed or uniformly random linear map, which is an autoencoding task, and the reported FID competitiveness does not establish that a diffusion model has been built or that the quantum components are doing generative modeling. Since the abstract's central claim rests on the model being a quantum diffusion model, this is the load-bearing weakness. The paper's own statement in Section 4.2 that 'an arbitrary noising transformation can be used' strengthens the concern rather than resolving it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes fully quantum, hybrid, and fully classical denoising diffusion models for CMS quark and gluon jet images. In the forward process, Gaussian noise is replaced by a single Haar-random unitary applied once per encoded channel; the denoiser is a U-Net with variational quantum circuit layers. Experiments on 1,000 16x16 ECAL jet images report final FID values of 1.8169 (classical), 1.8123 (hybrid), and 2.7362 (fully quantum), and the authors conclude that the quantum models are competitive with the classical model.","tokens_in":5919,"tokens_out":6303,"duration_ms":67944,"significance":"If the central claims hold, this would be a useful empirical demonstration of quantum circuits in a generative model for LHC data: the paper provides an open-source code repository and a direct comparison of classical, hybrid, and fully quantum variants on a physics dataset. However, the significance is currently limited by the absence of a derivation connecting the one-shot unitary forward process to a diffusion objective, by the lack of statistical uncertainty in the central FID comparison, and by the absence of any runtime or resource measurements supporting the stated computational motivation.","major_comments":[{"comment":"The quantum forward process is never defined as a conditional distribution q(x_t|x_0). Replacing Gaussian noise with Haar-random unitaries is not a noising operation unless a measurement is specified; if the unitary is followed by a computational-basis measurement, q becomes a discrete distribution, and if it is not, the denoiser input is a quantum state rather than the classical image the U-Net receives. In neither case is the KL divergence D_KL(q(x_T|x_0) || p(x_T)) in Eq. (5) shown to be finite or meaningful, and the one-shot unitary does not reproduce the multi-step schedule of Eqs. (1)-(3). The citation to cold diffusion [29] does not resolve this, because cold diffusion requires a sequence of degradation operators with increasing information loss; a unitary is invertible and norm-preserving. As written, the model is closer to an autoencoder that inverts a random linear map, so the central claim that the paper builds a quantum diffusion model is unsupported.","section":"§4.2, Eq. (5)"},{"comment":"The empirical comparison rests on single FID numbers (1.8169, 1.8123, 2.7362) with no error bars, no number of random seeds, no held-out split, and no description of how many generated samples are used for FID or whether the evaluation set was seen during training. The fully quantum value is about 50% worse than the classical one, so the word “competitive” requires at least a statistical test or a predefined acceptance threshold. This is load-bearing because the abstract's claim is an empirical one.","section":"§5–§6, FID results"},{"comment":"The stated motivation is to mitigate the computational challenges of diffusion models, but the paper reports no wall-clock training time, no qubit counts or circuit depth, no simulation cost, and no comparison of quantum or hybrid training overhead against the classical baseline. Since all experiments are simulations, the concluding suggestion that computations can be offloaded to faster quantum processors to reduce training time is not supported by any measurement in the manuscript.","section":"§1, §6"},{"comment":"The paper does not compare against a non-diffusion baseline such as a classical autoencoder with the same U-Net. Since the one-shot unitary forward map is invertible, the reported FID could in principle be achieved by learning the inverse map plus training-set statistics; an autoencoder baseline would show whether the generative formulation is doing any work. This comparison is also needed to test the assertion in Section 4.2 that the choice of forward scrambling does not significantly impact model performance.","section":"§4.2, §6"}],"minor_comments":[{"comment":"The section heading “Data A vailability” contains a typo and an extra space; it should read “Data Availability.”","section":"§8"},{"comment":"The dataset description mentions ECAL, HCAL, and reconstructed tracks, but the experiments use only 1,000 ECAL-detected jets cropped to 16x16 pixels; please state explicitly whether the other subdetector channels are discarded and how the 1,000-jet subset is selected.","section":"§2.3"},{"comment":"The phrase “Haar measurement” is nonstandard: the Haar measure is a distribution over unitary matrices, not a measurement. Please clarify the intended terminology and define the operation precisely in the quantum forward process.","section":"§2.2"},{"comment":"Reference [26] is cited as an example of molecular biology applications, but the title concerns dental smile analysis; please correct the citation or adjust the example.","section":"§1, Refs. [4,26]"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is readable and addresses a topical problem, but it needs substantially more rigor before publication. The main risk is the undefined forward process: either the authors provide a proper derivation of the diffusion-style objective for unitary noising (including the measurement step), or they reframe the work as a quantum autoencoder and compare against appropriate classical baselines. I recommend major revision rather than rejection because the core idea could be salvaged with additional theory and evaluation, but the current text does not support the central claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this paper is a proof-of-concept for putting quantum circuits inside a denoising architecture for CMS jet images, with open code. But the 'quantum diffusion' framing is not supported. The one-shot Haar-unitary forward step is invertible and information-preserving, so q(x_T|x_0) is a delta function and the KL term in Eq. 5 is undefined. Cold diffusion [29] doesn't cover a single unitary; it relies on a sequence of degrading transforms. As written, the reverse net learns to invert a fixed random unitary: that is an autoencoder, not a diffusion model. That is the load-bearing weakness.\n\nThe paper does have real value. The application to quark/gluon jets is new: QuDDPM [15] was for quantum states, and the hybrid model [28] was demonstrated on MNIST. This is the first direct comparison of fully quantum, hybrid, and classical denoisers on a structured HEP dataset. The architecture is clearly described, the training curves and samples are shown, and the code is open. That is more than many QML papers provide.\n\nThe empirical support is thin. The FID numbers are single runs: 1.8169 classical, 1.8123 hybrid, 2.7362 fully quantum. No error bars, no seed counts, no description of the evaluation set. Calling the fully quantum result 'competitive' when it is 50% worse than the baseline overstates the case. The discussion about offloading to faster quantum processors is speculative without any timing or hardware data.\n\nAlso minor: the dataset is cropped to 1,000 jets at 16x16, so the benchmark is narrow.\n\nWho would profit: researchers in QML-for-HEP who want a concrete baseline to build on. But they should treat this as a quantum autoencoder benchmark until the diffusion claim is justified.\n\nFor peer review, I would send it out with a request for major revision: a derivation or ablation of the unitary noising step (or a reframing as an autoencoder), seeded repeats with uncertainty, and either hardware/timing or a softened claim. The idea is worth engaging with; the current framing is not.","headline":"Applies quantum/hybrid DDPM-style models to CMS jets with open code, but the one-shot unitary 'noising' makes it an autoencoder in disguise, and the single-run FID does not support the competitiveness claim.","tokens_in":6574,"tokens_out":4877,"would_cite":false,"duration_ms":47652,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that fully quantum and hybrid diffusion models generate quark and gluon jet images whose FID scores (2.7362 and 1.8123) are on par with a classical model's 1.8169.","keywords":["quantum diffusion model","variational quantum circuits","quark-gluon jets","generative modeling","random unitary scrambling","denoising U-Net","FID","quantum machine learning"],"falsifier":"Train this one-shot unitary quantum model and a conventional multi-step Gaussian diffusion model on identical 16x16 jet images, then measure whether the quantum model's samples remain distinguishable from the training set by a hold-out classifier and whether FID parity persists when the number of sampling steps and the amount of scrambling are varied; if the quantum model collapses to memorized or reconstructed examples, or if its samples are classified with near-perfect accuracy, the claim that it is a competitive diffusion model would be falsified.","tokens_in":5462,"feed_emoji":"⚛️","tokens_out":10030,"duration_ms":90769,"temperature":0.7,"pith_summary":"The paper claims that the expensive noising and denoising pipeline of a diffusion model can be partially or fully run through quantum circuits without losing generative quality, on a task that matters for high-energy physics: generating images of quark and gluon jets. Using 16x16 jet images derived from particle-collision detector data, the authors replace the usual multi-step Gaussian noise schedule with a single random unitary scramble applied to each of four encoded channels, and put a variational quantum circuit inside the U-Net denoiser. They report that the hybrid model reaches an image-quality score (FID) of 1.8123 and the fully quantum model a score of 2.7362, compared with 1.8169 for a structurally similar classical model. The paper's point is that quantum components can be swapped into a generative model for physics data at essentially no quality loss, which it reads as evidence that heavy parts of deep learning could be offloaded to future quantum processors. No runtime advantage is claimed; the experiments are simulations.","feed_headline":"Hybrid quantum model matches classical on jet images","feed_subtitle":"Fully quantum FID 2.74 and hybrid FID 1.81 versus a classical baseline of 1.82 on quark and gluon jets.","key_machinery":"The load-bearing object is a one-shot random unitary scrambling step replacing the Gaussian noising process. A random unitary matrix is a reversible quantum operation, and sampling it uniformly from the space of quantum operations gives a scramble that is meant to destroy the original pixel structure the way repeated Gaussian noise would. The classical-to-quantum embedding uses $R_x$ rotation gates that encode groups of four pixels into four channels, and the denoising side uses a variational quantum circuit built from rotation and strongly entangling gates, kept identical across models, inserted inside a U-Net. The same circuit, with the number of layers as a tunable parameter, serves as the sole denoiser in the fully quantum model and as the middle layer of a convolutional U-Net in the hybrid model. Training minimizes the mean squared error between the model output and the clean image, which is the simplified diffusion objective.","core_discovery":"The central claim, stated on the paper's own terms, is empirical: a fully quantum diffusion model and a hybrid quantum-classical diffusion model generate quark and gluon jet images whose quality is competitive with a similar classical diffusion model. The forward diffusion process is changed from adding Gaussian noise over many timesteps to applying one random unitary matrix, sampled uniformly from the space of quantum operations, to each of four channels of an angle-encoded jet image; the denoising network is a U-Net whose core contains a variational quantum circuit with strongly entangling layers. After 50 epochs of training with mean squared error loss and the Adam optimizer, the final FID values are 1.8123 for the hybrid model, 2.7362 for the fully quantum model, and 1.8169 for the classical baseline. The authors therefore conclude that all or part of the deep neural network computations can in principle be moved to quantum processors without a performance trade-off. The paper does not claim a speedup; it claims parity in output quality.","pith_inferences":["Editorial extension: because a unitary is invertible and preserves information, one application does not progressively destroy data the way Gaussian noise does; the reported FID parity may therefore be evidence of autoencoder-like reconstruction rather than diffusion sampling, a distinction the paper does not test.","Editorial extension: a direct comparison with a classical one-shot denoiser trained on the same single-step scramble would isolate whether the quantum circuit adds capability beyond a classical network of comparable size.","Editorial extension: a sharper evaluation would measure sample diversity and mode coverage, for example with a hold-out classifier trained to distinguish generated jets from training jets, since FID alone can miss memorization or collapse.","Editorial extension: because the experiments use simulators, the FID parity does not by itself imply any computational saving; the practical value depends on the speed and cost of future quantum hardware."],"forward_implications":["If the parity result holds, quantum circuits can be placed inside the denoising stage of a diffusion model for physics data without a measurable drop in FID.","The one-shot unitary forward process removes the need for many noising timesteps, because the random scramble is applied once per channel rather than iteratively.","The single generated channel should extend to the full set of three subdetector channels per jet, a direction the paper explicitly leaves for future work.","Moving the quantum layers to real quantum processors could reduce training time while keeping generative quality, though the paper supports this only as an implication, not by a runtime measurement.","The FID plateau across all models suggests that data sparsity is the limiting factor, so keeping only the most prominent pixel values in a post-processing step could sharpen the generated jets."],"supporting_citations":[{"why":"introduces the idea of scrambling quantum states with random unitaries as the forward noising process and provides the QuDDPM baseline this paper adapts.","marker":"[15]"},{"why":"supplies the finding that the choice of forward scrambling does not significantly affect performance, which justifies using a one-shot arbitrary unitary instead of Gaussian noise.","marker":"[29]"},{"why":"proposes a hybrid quantum-classical diffusion U-Net on MNIST that this paper extends to the more complex quark-gluon data.","marker":"[28]"},{"why":"defines the image-quality metric FID used to compare generated and real jet images across all three models.","marker":"[27]"},{"why":"provides the open detector dataset and its preprocessing, including the cropping to 16x16 images used for training.","marker":"[25]"},{"why":"supplies the quantum-circuit simulator used to run and train the quantum models in the absence of quantum hardware.","marker":"[11]"}],"fun_headline_variants":["Quantum diffusion rivals classical for jet generation","Fully quantum diffusion model matches classical on jets","Hybrid quantum diffusion competes with classical on jets","Quantum diffusion model rivals classical on jet images","Quantum diffusion for jets: competitive with classical"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a single random reversible quantum operation applied to each channel behaves like the multi-step noise process of a diffusion model, an assumption the paper adopts from earlier work on arbitrary image transforms; if it fails, the competitive scores may reflect autoencoder-like reconstruction rather than genuine diffusion sampling.","fun_headline_variants_meta":{"raw":{"variants":["Quantum diffusion rivals classical for jet generation","Fully quantum diffusion model matches classical on jets","Hybrid quantum diffusion competes with classical on jets","Quantum diffusion model rivals classical on jet images","Quantum diffusion for jets: competitive with classical"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000698,"raw_usage":{"total_tokens":3115,"prompt_tokens":870,"completion_tokens":2245,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":486,"completion_tokens_details":{"reasoning_tokens":2191}},"tokens_in":486,"tokens_out":2245,"duration_ms":15848,"temperature":1.0,"reasoning_tokens":2191,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:03:46.814551+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train this one-shot unitary quantum model and a conventional multi-step Gaussian diffusion model on identical 16x16 jet images, then measure whether the quantum model's samples remain distinguishable from the training set by a hold-out classifier and whether FID parity persists when the number of sampling steps and the amount of scrambling are varied; if the quantum model collapses to memorized or reconstructed examples, or if its samples are classified with near-perfect accuracy, the claim that it is a competitive diffusion model would be falsified.","supporting_citations":[{"cited_title":"Generative quantum machine learning via denoising diffusion probabilistic models","cited_arxiv_id":null,"evidence_quote":"introduces the idea of scrambling quantum states with random unitaries as the forward noising process and provides the QuDDPM baseline this paper adapts."},{"cited_title":"Cold diffusion: Inverting arbitrary image transforms without noise","cited_arxiv_id":null,"evidence_quote":"supplies the finding that the choice of forward scrambling does not significantly affect performance, which justifies using a one-shot arbitrary unitary instead of Gaussian noise."},{"cited_title":"Burkle, S","cited_arxiv_id":null,"evidence_quote":"provides the open detector dataset and its preprocessing, including the cropping to 16x16 images used for training."}],"review_version":1}