betaFeedback

WorksPhysics and Astronomy

Machine-learning-based neutrino-event generation constrained by experimental measurements

Abstract

We present a generative surrogate for charged-current νμν_μ-nucleus scattering that can be calibrated to measured differential cross sections without event-level experimental samples. Starting from a theoretical interaction model, we fine-tune the surrogate to a MicroBooNE measurement and show that this also improves the description of observables not used in the fit. The calibrated model can also be applied to measurements from different neutrino fluxes without retuning, while retaining event-level predictions of outgoing-muon kinematics. Our results show that experimental cross sections can directly constrain generative event models while preserving their applicability across different observables and beam configurations.

Introduction. Precision measurements of neutrino oscillations and leptonic CP violation are demanding tests for establishing the Standard Model’s consistency. They may also shed light on the matter-antimatter asymmetry of the Universe [1]. To achieve these precision measurements at next-generation long-baseline neutrino experiments, such as DUNE [2] and Hyper-Kamiokande [3], an accurate modeling of neutrino–nucleus interactions is required. Nuclear effects matter because they affect final states and neutrino-energy reconstruction [4, 5]. Recent studies show that uncertainties in final-state interaction modeling can induce distortions comparable to the precision required for oscillation-parameter measurements [6]. At the same time, increasingly sophisticated theoretical neutrino interaction models make efficient event generation in Monte Carlo (MC) simulation tools increasingly challenging [11, 12, 13] [7, 8, 9, 10]. Generative deep-learning models provide natural surrogates for stochastic particle–nucleus scattering, learning multidimensional distributions of final-state observables and allowing efficient event sampling [14, 15, 16, 17, 18]. Beyond reproducing a single process, transfer learning can extend such models to related processes [19].

Accelerator neutrino beams are not monoenergetic but have a broad, energy-dependent flux [20]. Measured cross sections are flux-averaged, with several reaction mechanisms contributing simultaneously, complicating both model validation and MC event-generator tuning. For a fixed nuclear target, however, interaction physics, nuclear effects, and cross section distributions depend on (Eν,flavor)(E_\nu,\mathrm{flavor}), independently of how the neutrinos were produced. Detector response, acceptance, and backgrounds are experiment-specific and enter the measurement and its uncertainties. At the interaction level, however, experiments using the same nuclear target differ primarily through their neutrino flux Φ(Eν,flavor)\Phi(E_\nu,\mathrm{flavor}). A surrogate conditioned on EνE_\nu can therefore transfer interaction information between beam configurations, with the beam profile as the experiment-specific input.

We realize this with a generative adversarial network (GAN) which readily learns unnormalized event distributions for neutrino and antineutrino scattering off nuclei [17, 18], but an unnormalized shape cannot be compared to a measured cross section nor transported to a different beam. The missing ingredient is the energy-dependent interaction probability. This weighting is precisely what a bare generator lacks. We therefore couple the GAN to a deep neural network (DNN) that predicts the total cross section σ(Eν)\sigma(E_\nu), so we can normalize generated events to the cross section. The surrogate can then be fine-tuned directly on flux-averaged measurements, and its interaction-level predictions can be transported to other experiments by supplying the corresponding neutrino flux, enabling a direct comparison between the surrogate and measurements not used in the training. In this sense, the framework provides a way to fine-tune an MC interaction model directly from the measured cross sections, without requiring event-level experimental information. This differs from machine-learning unfolding approaches such as OmniFold [21], which iteratively reweight simulated events using detector-level data and detector simulation. Instead, our approach modifies the generative model and is constrained directly by binned cross sections and their covariance, rather than per-event detector-level samples. It also differs from the parametric generator tunes [22], which adjust a predefined set of parameters of the underlying interaction model.

The central idea is to retain the event-level structure learned from a physics-based MC model while allowing measured cross sections to modify its predictions. We therefore couple a σ\sigma-DNN, which supplies the energy-dependent normalization, to a conditional GAN that generates the outgoing-muon kinematics (Eμ,θ)(E_\mu,\theta) at fixed EνE_\nu. Relative to Ref. [17], the key extensions are the explicit normalization for arbitrary neutrino fluxes and direct fine-tuning to measured cross sections rather than simulated events alone. The resulting surrogate therefore combines MC-based event structure with direct experimental constraints and can be applied across different beam configurations.

We test this strategy by calibrating the model to a MicroBooNE measurement [23] of the inclusive charged-current (CC) νμ\nu_\mu-argon scattering double-differential cross section d2σ/dpμdcos⁡θd^2\sigma/dp_\mu d\cos\theta, and evaluating it against measurements involving different observables and neutrino fluxes. The tuned surrogate improves the description of MicroBooNE observables not used in the training [24, 25, 26] and is consistent with the ArgoNeuT measurements [27, 28] obtained with a substantially different flux. These results show that experimentally measured cross sections can constrain a generative event model while remaining applicable across different observables and beam configurations.

We construct our model in two stages. First, we obtain the σ\sigma-DNN and GAN models independently from data produced with the NuWro MC event generator. Then we fine-tune both models simultaneously using experimental data. Figure 1 illustrates the model optimization scheme.

Figure 1. Training scheme: (a) The σ\sigma-DNN and GAN are pre-trained on NuWro. (b) Fine-tuning alternates adversarial regularization using the data-informed reservoir with a differentiable full-covariance fit to the forward-smeared MicroBooNE measurement. The latter updates GG and the σ\sigma-DNN.

The first stage. The σ\sigma-DNN is a feed-forward fully connected network. Rather than mapping a single energy to a cross section, it predicts the cross section averaged over an energy interval, with the single-energy value recovered for a vanishing interval width.

Using the σ\sigma-DNN and a known experimental flux Φ\Phi (in our case, from MicroBooNE), we obtain a sample of interacting neutrino energies. Namely, the total-cross-section-weighted energy distribution of incoming neutrino events reads: fi=σ^iΦi/⟨σ^Ω⟩f_i = \hat{\sigma}_i\Phi_i/\langle\hat{\sigma}_{\Omega}\rangle, where σ^i\hat{\sigma}_i is the total cross section predicted by the σ\sigma-DNN (or by NuWro) (averaged over the ii-th bin), and ⟨σ^Ω⟩=∑i∈Ωσ^iΦi\langle\hat{\sigma}_{\Omega}\rangle= \sum_{i\in\Omega}\hat{\sigma}_i\Phi_i, is the flux-averaged cross section, where the flux satisfies: ∑i∈ΩΦi=1\sum_{i\in\Omega}\Phi_i = 1, and Ω\Omega is the domain covered by the flux.

To generate events, we use a GAN conditioned on the neutrino energy EνE_\nu to reproduce the muon kinematics as a function of EνE_\nu. The generator GG maps a latent noise vector, together with EνE_\nu, to the muon energy and scattering angle (Eμ,θ)(E_\mu,\theta); the discriminator DD learns to distinguish these generated events from NuWro training events. We trained the network from scratch on CC νμ\nu_\mu-argon events generated with NuWro in the same manner as in Ref. [18]. Technical details of the σ\sigma-DNN, GAN architecture and training, fine-tuning procedure, differentiable histogramming, and additional comparisons are given in the Supplemental Material (SM) [29].

Using the event distribution fif_i, we randomly sample neutrino energies for the GAN generator to obtain cross-section-weighted event samples. Differential cross sections are then reconstructed from the generated event fractions and the corresponding σ\sigma-DNN normalization; the explicit construction used in the fine-tuning is given in the SM.

In the second stage, we fine-tune the σ\sigma-DNN and GAN to the MicroBooNE CC νμ\nu_\mu-argon measurement d2σ/dpμdcos⁡θd^2\sigma/dp_\mu d\cos\theta [23]. We achieve this by fine-tuning the models using a nonstandard procedure. The difficulty

is twofold: the MicroBooNE data are presented as a three-dimensional differential cross-section distribution (double-differential in bins of EνE_\nu), whereas the GAN is trained on event samples; and the differential distributions also encode the total cross section, which in our system is reproduced by the σ\sigma-DNN. Both observations lead us to a joint optimization procedure in which the GAN and σ\sigma-DNN weights are updated together using the experimental distributions.

The second-stage optimization combines two complementary sub-steps. In the first, the discriminator is fine-tuned on events drawn from a reservoir, while the generator is updated adversarially against the discriminator; only the final layers of both networks are allowed to adapt. The generator samples neutrino energies from the flux reweighted by the current σ\sigma-DNN prediction.

The reservoir consists of one million truth-level NuWro events resampled from a larger sample generated with the MicroBooNE flux. The MicroBooNE result is a smeared double-differential cross section in (pμ,cos⁡θμ)(p_\mu,\cos\theta_\mu), reported in bins of EνE_\nu, with the smearing matrix AcA_c relating the truth-level prediction to the measured bin space. The measured bin contents, together with the MicroBooNE flux, determine non-negative target occupancies for the reservoir, as detailed in the SM. Since the reservoir is constructed from truth-level NuWro events using a smeared measurement, it cannot be interpreted as a truth-level target; rather, it regularizes the generator toward the physical, NuWro-like manifold. We perform the quantitative fit separately in the second sub-step, where we forward-smear the generated prediction with AcA_c and compare it with the original measurement.

In the second sub-step, we update the generator and the σ\sigma-DNN jointly using the differential cross-section measurement. We histogram events generated by GG, and normalize each histogram by the σ\sigma-DNN prediction for the flux-averaged cross section ⟨σ^ω⟩\langle\hat{\sigma}_{\omega}\rangle over the corresponding neutrino-energy interval ω\omega. This yields an estimate of the double-differential cross section in each EνE_{\nu} interval. The prediction is then forward-smeared, s^ω=Acσ^ω\hat{s}_{\omega}=A_c\hat{\sigma}_{\omega}, and compared with the measured sωs_{\omega} through a χc2\chi_c^2 constructed with the full covariance matrix; see Eq. S-IV.13 of the SM.

The full covariance is retained because the unfolding and the smearing AcA_c induce bin-to-bin correlations. If we do not follow this prescription, the fitting procedure collapses, as discussed in Ref. [30]. To make the comparison differentiable with respect to the network weights, the histogramming step is replaced by a smooth approximation, detailed in Sec. S-IV B of the SM. The sampling of EνE_{\nu} conditions the generation but does not contribute to the gradient; the normalization through ⟨σ^ω⟩\langle\hat{\sigma}_{\omega}\rangle does.

χv2/n\chi_v^2/nχc2/n\chi_c^2/n
σ\sigma-DNNB_B, GANB_B461.7/138460.3/138
σ\sigma-DNNB_B, GANT_T435.6/138268.4/138
σ\sigma-DNNT_T, GANB_B152.0/138500.1/138
σ\sigma-DNNT_T, GANT_T110.5/138220.3/138

Table I. The values of χv2\chi_v^2 and χc2\chi_c^2 over the number of bins (n) for four configurations of the models: base (subscript B) and fine-tuned (subscript T) σ\sigma-DNN and GAN models. Computed for samples of 1 million events.

Table I summarizes the effectiveness of the fine-tuning, while Figure 2 illustrates the resulting agreement with the data. We report two complementary metrics: the diagonal χv2\chi_v^2, defined in Eq. (S-IV.14) of the SM, which primarily probes per-bin agreement, and the full-covariance χc2\chi_c^2, which also accounts for the bin-to-bin correlations induced by the unfolding and the smearing matrix AcA_c. The table shows these metrics for the four possible post-training combinations of the base and tuned σ\sigma-DNN and GAN components.

Figure 2. Base and tuned models compared with the MicroBooNE double-differential cross-section measurement [23] for various EνE_{\nu} ranges. Grey vertical dashed lines separate different EνE_{\nu} regions, from left to right: [0.2, 0.705], [0.705, 1.05], [1.05, 1.57], and [1.57, 4] GeV.

Replacing only the base σ\sigma-DNN with its tuned counterpart lowers χv2\chi_v^2 sharply but does not improve χc2\chi_c^2, indicating that this component mainly corrects the cross-section normalization. Conversely, replacing only the base GAN with the tuned generator substantially improves χc2\chi_c^2 while leaving χv2\chi_v^2 nearly unchanged, showing its dominant role in reproducing correlated kinematic structure. Only the combination of both tuned components improves both

metrics simultaneously, demonstrating their complementary roles and motivating their joint optimization.

We note that the first sub-step, in which the discriminator is exposed to the reservoir, is essential. Without it, training with the χc2\chi_c^2 loss term alone reproduces the smeared, flux-averaged distributions but distorts the quasielastic (QE) and Δ\Delta-resonance peaks; see Figure 3: the smearing AcA_c and the coarse measurement binning wash out this sharp structure in the observable to which χc2\chi_c^2 is fit, so the covariance objective alone cannot constrain it. The reservoir, built from truth-level NuWro events that carry this structure, supplies exactly the differential information the smeared χc2\chi_c^2 lacks, keeping the generator on the physical manifold of reaction events.

Figure 3. Event distribution at Eν=3E_{\nu}=3 GeV generated by the fine-tuned GAN with and without the reservoir sub-step.

We next test whether the tuned surrogate is consistent with other published MicroBooNE cross-section measurements not used in the fit. We first consider an earlier MicroBooNE measurement [24] of the total cross section σ(Eν)\sigma(E_{\nu}) and the differential cross sections dσ/dEμd\sigma/dE_{\mu} and dσ/dEhadd\sigma/dE_{\mathrm{had}}, with Ehad=Eν−EμE_{\mathrm{had}}=E_{\nu}-E_{\mu}. The model was tuned only on the cross section of [23]; the observables considered here are therefore an out-of-training test of what the surrogate has learned, not a refit. Figure 4 shows the corresponding cross sections. The tuned model improves the description of all three observables relative to the base model (both in the shape and in the normalization): in the full-covariance metric χc2/n\chi_c^2/n improves from 58.1/10 to 44.2/10 for σ(Eν)\sigma(E_\nu), from 29.1/11 to 12.7/11 for dσ/dEμd\sigma/dE_\mu, and from 17.1/8 to 4.4/8 for dσ/dEhadd\sigma/dE_{\mathrm{had}}. The improvement is most pronounced at neutrino energies below 2 GeV and muon energies below 1 GeV, the region best constrained by the data.

Image not extracted

Figure 4. Base and fine-tuned models compared to the MicroBooNE measurement [24]: σ/⟨Eν⟩\sigma/\langle E_\nu\rangle (left), dσ/dEμd\sigma/dE_\mu (center) and dσ/dEhadd\sigma/dE_{\mathrm{had}} (right).

Finally, we compare our model against MicroBooNE and ArgoNeuT data not used in training. We start from more recent MicroBooNE measurements [25, 26] of dσ/dcos⁡θd\sigma/d\cos\theta cross section shown in Figure 5. The comparison for the double-differential cross section d2σ/dEμ dcos⁡θd^2\sigma/dE_\mu\,d\cos\theta [25, 26] is shown in SM. These measurements were not used at any stage of training or model selection; they therefore constitute a cross-measurement test of the surrogate. The tuned model improves markedly over the base model in both observables. In the full-covariance metric, χc2/n\chi_c^2/n drops from 257.0/69 to 109.2/69 for d2σ/dEμ dcos⁡θd^2\sigma/dE_\mu\,d\cos\theta, and from 291.5/17 to 6.8/17 for dσ/dcos⁡θd\sigma/d\cos\theta. The latter is a notable result: a model tuned on a single, distinct measurement predicts a later measurement not used in training at χc2/n<1\chi_c^2/n < 1, without any retuning. This shows that the data-driven adaptation

Image not extracted

Figure 5. Base model, tuned model, and the MicroBooNE data [25, 26] for dσ/dcos⁡θd\sigma/d\cos\theta, not used in the training.

remains consistent with a different representation of the same underlying interaction kinematics.

ArgoNeuT provides a qualitatively different test: it uses the NuMI beam in its low-energy configuration [31], whose flux extends well above the energies covered by the training data, and it is a distinct detector. We therefore use it primarily to probe the extrapolation of the framework rather than the data-driven adaptation within the constrained region. We obtain predictions with NuWro and our DNN using the fluxes of Refs. [27, 28]. ArgoNeuT did not publish the covariance matrix, so a full χc2\chi_c^2 cannot be formed; we report the diagonal χν2\chi_\nu^2 only as an indicator, noting that it neglects the bin-to-bin correlations of the measurement.

Figure 6 shows the dσ/dpμd\sigma/dp_\mu distribution for ArgoNeuT; see also the SM for the antineutrino-mode results. Because the ArgoNeuT flux lies almost entirely above the energy range constrained by the MicroBooNE fine-tuning, the tuned and base models give similar flux-averaged predictions, and the agreement with the data is therefore inherited mainly from the base model’s high-energy behavior. Both models yield χν2/n≲1\chi_\nu^2/n \lesssim1, although this partly reflects the large ArgoNeuT uncertainties. Within these limitations, the tuned surrogate remains consistent with the out-of-training ArgoNeuT data, indicating that the fine-tuning does not substantially degrade the high-energy behavior inherited from the base model.

Image not extracted

Figure 6. Base model, tuned model, and ArgoNeuT data [27, 28] for dσ/dpμd\sigma/dp_\mu distribution. The shaded area shows the pμp_\mu range of the training MicroBooNE data.

In summary, we have constructed a machine-learning-based neutrino-event generator that can be calibrated directly to binned experimental cross sections without event-level experimental samples. A conditional GAN generates the outgoing-muon kinematics (energy EμE_{\mu} and scattering angle θ\theta), while a σ\sigma-DNN provides the energy-dependent absolute normalization; both are jointly constrained by the forward-smeared MicroBooNE measurement. The central result is that experimental data can update a generator pretrained on a conventional Monte Carlo model, and it can then be used for different observables and flux configurations.

A model tuned to the MicroBooNE measurement improves the description of other MicroBooNE observables not used in training and remains consistent with ArgoNeuT data obtained with a substantially different flux. The data-constrained region modifies the normalization and correlated kinematics, while outside this region the prediction largely inherits the underlying MC model. The framework therefore defines a route toward event generators that combine theory-based simulations with direct experimental constraints; extending the constrained kinematic range and propagating calibrated model uncertainties are natural next steps. With the latter in place, the surrogate could serve as a common interaction reference that, folded with each beam’s flux, tests the mutual consistency of cross-section measurements across experiments.

The fine-tuned models and the event-generation script are available in Ref. [32].

Acknowledgments

This research is partly (K.M.G., B.E.K., J.T.S.) or fully (J.L.B., H.P., R.D.B.) supported by the National Science Centre under grant UMO-2021/41/B/ST2/02778. J.L.B. and J.T.S. are also supported by the Polish Ministry of Science Grant No. 2022/WK/15.

Model optimization

In the following sections, we describe in detail all steps of the model optimization. Figure S1 presents an expanded version of the model-optimization scheme shown in Figure 1 of the main text.

Figure S1. Training scheme. (a) The σ\sigma-DNN and the conditional GAN (GG, DD) are first trained on NuWro. (b) Fine-tuning combines (1) adversarial regularization using the data-informed reservoir and (2) a differentiable full-covariance fit in the measurement space. The discriminator is updated on every reservoir batch, while every fifth discriminator update is followed by an adversarial update of GG and by the χc2\chi_c^2 update of GG and the σ\sigma-DNN. Sampling of EνE_\nu from Φσ^\Phi\hat{\sigma} carries no gradient.

Cross-section DNN: parametrization and training

The σ\sigma-DNN is a feed-forward network that supplies the interaction cross section used to weight the events generated by the GAN. Beyond a prediction at a fixed neutrino energy, it must return the cross section averaged over an energy interval, so that flux-averaged cross sections can be evaluated over the measurement bins. We describe in turn the tabulated training data, the input and target parametrization, the architecture, and the optimization.

Training data and interval averaging

We first tabulate cross-section values with NuWro, sampling neutrino energies from 0 to 20000 MeV in steps of 5 MeV,

Eν,i∈{0,5,…,19995,20000} (MeV),(1)E_{\nu,i} \in\{0, 5, \ldots, 19995, 20000\}\ \text{(MeV)}, \tag*{(1)}

with the corresponding cross sections

σi∈{σ(0),σ(5),…,σ(19995),σ(20000)}(2)\sigma_i \in\{\sigma(0),\sigma(5),\ldots,\sigma(19995),\sigma(20000)\} \tag*{(2)}

in units of cm2\mathrm{cm}^2. Below the reaction threshold, the cross section is set to zero. To obtain the average cross section over an energy interval ΔEν\Delta E_\nu (a bin width, which in our analysis reads 50 MeV), two table entries with indices m,n∈Nm,n \in\mathbb{N}, n≤mn \le m, corresponding to Eν,nE_{\nu,n}, Eν,mE_{\nu,m} and σn\sigma_n, σm\sigma_m, are selected at random, and the cross section is averaged over all tabulated values within the interval,

⟨σmn⟩=1m−n+1∑i=nmσi,(3)\langle\sigma_{mn} \rangle= \frac{1}{m-n+1}\sum_{i=n}^{m}\sigma_i, \tag*{(3)}

which is the flux-averaged cross section under the assumption of a uniform flux across the interval; this local-uniformity assumption is what motivates the finer flux subdivision Φ→Φ′\Phi\to\Phi' used in following sections. The single-energy prediction is recovered as the special case n=mn=m.

Input and target parametrization

Since the cross section is unbounded, growing approximately linearly at large neutrino energies, the network is trained not on ⟨σmn⟩\langle\sigma_{mn} \rangle directly but on the bounded target

σmn′=⟨σmn⟩×1038Emn4,Emn=Eν,m+Eν,n2×10−3,(4)\sigma'_{mn} = \sqrt[4]{\frac{\langle\sigma_{mn} \rangle\times10^{38}}{E_{mn}}}, \qquad E_{mn} = \frac{E_{\nu,m}+E_{\nu,n}}{2}\times10^{-3}, \tag*{(4)}

where dividing by EmnE_{mn} bounds the quantity from above, and after rescaling EmnE_{mn} is in GeV units. The physical cross section is recovered by inverting Eq. S-II.4 after prediction. The energy inputs are rescaled to the unit interval,

Eν,m′=Eν,m2×104,Eν′∈[0,1],(5)E'_{\nu,m} = \frac{E_{\nu,m}}{2\times10^4}, \qquad E'_\nu\in[0,1], \tag*{(5)}

and the input for each sample is the tuple

input=(Eν,m′,Eν,n′,Eν,m′−Eν,n′).(6)\mathrm{input} = (E'_{\nu,m}, E'_{\nu,n}, E'_{\nu,m}-E'_{\nu,n}). \tag*{(6)}

To improve convergence, the interval width is constrained during training by Eν,m−Eν,n≤1000E_{\nu,m}-E_{\nu,n} \le1000 MeV, so the model is trained only on intervals narrower than 1 GeV.

Architecture

The network is built from seven blocks, each consisting of a dense layer followed by layer normalization and a hyperbolic-tangent activation. The blocks contain 512, 256, 128, 64, 32, 16, and 8 neurons, respectively. A final dense layer produces a single output, activated by a softplus function

f(x)=ln⁡(1+e10x)10,(7)f(x) = \frac{\ln(1+e^{10x})}{10}, \tag*{(7)}

whose lower bound guarantees a non-negative predicted cross section; the factor of 10 controls the sharpness of the softplus knee.

Optimization

Three choices act together to keep the training target σmn′\sigma'_{mn} strictly positive. First, the softplus output cannot be negative. Second, to avoid zeros in the tabulated cross sections, we apply

⟨σmn⟩→⟨σmn⟩+ϵ,if ⟨σmn⟩=0,(8)\langle\sigma_{mn} \rangle\to\langle\sigma_{mn} \rangle+\epsilon,\quad\text{if }\langle\sigma_{mn} \rangle=0, \tag*{(8)}

with ϵ=14×10−48 cm2\epsilon= 14 \times10^{-48}\,\mathrm{cm}^{2}. Third, the loss is normalized by the target itself,

L=1N∑mn∈batch(σmn′−σ^mn′)2σmn′,(9)L = \frac{1}{N} \sum_{mn \in\mathrm{batch}} \frac{(\sigma'_{mn} - \hat{\sigma}'_{mn})^{2}}{\sigma'_{mn}}, \tag*{(9)}

where σ^mn′\hat{\sigma}'_{mn} is the network prediction. This target-normalized form prevents the large-cross-section region from dominating the loss, so that the low-energy, near-threshold region, where σ\sigma is small, remains well fit; it also requires σmn′>0\sigma'_{mn} > 0, which the two preceding steps ensure. We train with the Nadam optimizer [33] at a learning rate of 0.001, using mini-batches of size N=1000N = 1000. The single-energy result (m=nm = n) is shown in Figure S2.

Image not extracted

Figure S2. Cross section predicted by NuWro and by the base neural network. Predictions were made on the energies sampled for training (NuWro) and points in the middle of the energies of the tables (DNN).

The trained σ\sigma-DNN is subsequently used as a flux reweighter. For each flux-energy bin it provides the cross-section average defined in Eq. (S-II.3), which determines the event-energy sampling probability used in the main text. Figure S2 validates the resulting single-energy prediction against the NuWro table.

Generative adversarial network: architecture and training

A generative adversarial network (GAN) consists of two networks trained in competition: a generator GG and a discriminator DD. The discriminator learns to distinguish real samples, drawn from the training set, from fake samples produced by GG, while GG learns to produce samples that DD judges to be real. At convergence, GG produces samples indistinguishable from the real ones. GANs for νμ\nu_{\mu} scattering and their domain adaptation via transfer learning were discussed in Refs. [17, 18]. We use a conditional GAN (cGAN), since we reproduce the event kinematics as a function of the neutrino energy EνE_{\nu}.

Implementation and building blocks

We implement the network in Keras 2.15 [34]. The basic feature-extraction block combines a dense layer, layer normalization [35], an activation, and a dropout layer. The activations are ReLU [36] for GG and PReLU [37] for DD. Zero-centered Gaussian noise with learnable parameters is injected into the inputs and after each layer (except for the first dense layer that receives the noisy inputs), at a rate of 0.0001 for GG and 0.001 for DD.

The generator receives a 50-dimensional latent vector drawn from N(0,1)\mathcal{N}(0, 1) together with the rescaled neutrino energy. Each of the two inputs is processed by its own dense layer with 100 neurons, then concatenated, and dropout is applied, and passed to the main body of GG. The generator outputs a pair of values representing the muon kinematic variables, activated by a hyperbolic tangent that constrains them to (−1,1)(-1, 1). The discriminator receives the tuple (Eμ′,θ′)(E'_{\mu}, \theta') together with Eν′E'_{\nu}, from either the training set or a generated sample, which are concatenated and passed to a dense layer with 282 neurons, then dropout, and finally to the main body of DD. The main body of GG comprises three consecutive blocks of 100 neurons each; the body of DD comprises six blocks of 141 neurons each, so that each DD block carries roughly twice the number of parameters of a GG block.

Variable parametrization

The networks do not process the physical variables EνE_\nu, EμE_\mu, θ\theta directly. Instead, we use the transformations

Eν′=ln⁡[Eν  (GeV)+eEν,min⁡  (GeV)+e]−1,E_\nu^{\prime}=\ln\left[\frac{E_\nu\;(\mathrm{GeV})+e}{E_{\nu,\min}\;(\mathrm{GeV})+e}\right]-1,
Eμ′=21−1−1−1−Eμ−mμEν−mμ−1,(10)E_\mu^{\prime}=2\sqrt{1-\sqrt{1-\sqrt{1-\sqrt{1-\frac{E_\mu-m_\mu}{E_\nu-m_\mu}}}-1}}, \tag*{(10)}
θ′=21−1−1−1−θ  (rad)/π−1.\theta^{\prime}=2\sqrt{1-\sqrt{1-\sqrt{1-\sqrt{1-\theta\;(\mathrm{rad})/\pi}}}}-1.

with Eν,min⁡=0.11E_{\nu,\min}=0.11 GeV (the lowest neutrino energy in the training NuWro dataset is Eν≈0.112E_\nu\approx0.112 GeV) and ee is the base of the natural logarithm. Training on the muon kinematics proved difficult, especially at low energies Eν<1E_\nu<1 GeV, which motivates this choice: the parametrization (empirically tuned to flatten the distribution near the phase-space boundaries) constrains the outputs to physically valid events; the muon can be neither more energetic than the incoming neutrino nor less energetic than its rest mass, and it pushes the kinematic distributions away from the edges of the allowed phase space. As a result, GG need not drive its weights to large values to reproduce the distribution, while DD can still resolve its tails.

Dense skip connections

Because adversarial training is inherently unstable, we make heavy use of skip connections to ease the flow of information through both networks. The output of each block is made available as an input to every subsequent block, yielding densely connected networks; the inputs to each block are averaged over their number, as shown in Figure S3. Such architectures have been explored extensively, e.g. residual networks [38] and densely connected networks [39].

Figure S3. The output of each block of GG and DD is used as an input to all subsequent blocks; all inputs to a block are averaged before it is processed.

Training

The GAN was trained from scratch on a CC νμ\nu_\mu–argon sample generated with NuWro, to reproduce the muon energy Eμ′E_\mu^{\prime} and scattering angle θ′\theta^{\prime} distributions. The training set contains 20 million events with neutrino energies in the range ∼112\sim112–20000 MeV. The spectrum is deliberately irregular, with more events generated at low energies than at high ones, and the same spectrum is used to feed GG; this prevents DD from learning the energy spectrum instead of the muon kinematics.

Each discriminator batch consists of 2000 training events plus 2000 from GG, while each generator batch consists of 4000 events; one generator step is taken for every five discriminator steps. The model was trained for several hundred epochs. The learning rate was initialized at 10−410^{-4} for D and 10−510^{-5} for G and reduced by a factor of ten in stages over the course of training. The optimizer was AdamW with amsgrad [40] enabled at the final stages of training.

Fine-tuning

Reservoir construction and the regularization sub-step

The fine-tuning must reconcile two mismatched representations: the GAN operates on individual events, while the MicroBooNE result [23] is a three-dimensional differential cross section d2σ/dpμ dcos⁡θd^{2}\sigma/dp_{\mu}\,d\cos\theta in bins of EνE_{\nu}. To supply the GAN with event-by-event information informed by the measurement, we construct a reservoir: a sample of NuWro events whose target occupancies in (Eν,pμ,cos⁡θμ)(E_{\nu},p_{\mu},\cos\theta_{\mu}) are determined from the measured cross section and the MicroBooNE neutrino flux.

From the measured cross section to reservoir occupancies

The MicroBooNE measurement is reported as a smeared cross section ss, related to a truth-level prediction σ^\hat{\sigma} by

s^=Acσ^,(11)\hat{s}=A_{c}\hat{\sigma}, \tag*{(11)}

where AcA_{c} is the smearing matrix provided with the data, acting on the measurement bins.

Let sωijs_{\omega ij} denote the measured double-differential cross section in the neutrino-energy region ω\omega, muon-momentum bin ii, and angular bin jj. To construct non-negative sampling probabilities for the reservoir, we define the bin weight

qωij=∣sωij∣Δpμ,i Δcos⁡θj.(12)q_{\omega ij}=\lvert s_{\omega ij}\rvert\Delta p_{\mu,i}\,\Delta\cos\theta_{j}. \tag*{(12)}

The absolute value is used only in constructing the reservoir because the regularized unfolding yields negative central values in some low-statistics bins, which cannot correspond to negative event counts. The original signed measurement is retained in the subsequent covariance-based fit.

For each neutrino-energy region, we define the corresponding integrated non-negative weight

Qω=∑ijqωij.(13)Q_{\omega}=\sum_{ij}q_{\omega ij}. \tag*{(13)}

Let

Fω=∫ωdEνΦ(Eν)∫ΩresdEνΦ(Eν),Ωres=⋃ωω(14)F_{\omega}=\frac{\int_{\omega}dE_{\nu}\Phi(E_{\nu})}{\int_{\Omega_{\mathrm{res}}}dE_{\nu}\Phi(E_{\nu})},\qquad\Omega_{\mathrm{res}}=\bigcup_{\omega}\omega \tag*{(14)}

denote the fraction of the MicroBooNE flux within Ωres\Omega_{\mathrm{res}} contained in region ω\omega, such that

∑ωFω=1.\sum_{\omega}F_{\omega}=1.

The relative population of neutrino-energy regions in the reservoir is then

P(ω)=FωQω∑ω′Fω′Qω′.(15)P(\omega)=\frac{F_{\omega}Q_{\omega}}{\sum_{\omega'}F_{\omega'}Q_{\omega'}}. \tag*{(15)}

Within a fixed neutrino-energy region, the measured cross section determines the relative population of the muon-kinematic bins,

P(i,j∣ω)=qωijQω.(16)P(i,j\mid\omega)=\frac{q_{\omega ij}}{Q_{\omega}}. \tag*{(16)}

Consequently, the probability assigned to a three-dimensional reservoir bin is

P(ω,i,j)=Fω∣sωij∣Δpμ,iΔcos⁡θj∑ω′i′j′Fω′∣sω′i′j′∣Δpμ,i′Δcos⁡θj′.(17)P(\omega,i,j)=\frac{F_{\omega}\lvert s_{\omega ij}\rvert\Delta p_{\mu,i}\Delta\cos\theta_{j}}{\sum_{\omega'i'j'}F_{\omega'}\lvert s_{\omega'i'j'}\rvert\Delta p_{\mu,i'}\Delta\cos\theta_{j'}}. \tag*{(17)}

For a reservoir containing NN events, the target occupancy of each bin is therefore Nωij≃NP(ω,i,j)N_{\omega ij}\simeq NP(\omega,i,j).

Drawing the reservoir

We start from a large NuWro sample generated with the MicroBooNE neutrino-energy spectrum. For each three-dimensional bin (ω,i,j)(\omega, i, j), NuWro events falling in that bin are selected according to the target occupancy defined by Eq. (S-IV.7). The selected events are then combined and shuffled to form the one-million-event reservoir. The reservoir therefore reproduces the flux-weighted event occupancies inferred from the measured cross section while retaining truth-level NuWro kinematics.

The reservoir is not a direct experimental target. In particular, the measurement used to construct it is smeared, whereas the selected NuWro events retain their truth-level kinematics. The reservoir therefore acts as a data-informed regularization sample; the quantitative fit to the measurement is performed separately in the second sub-step.

Figure S4. One-million-event reservoir used to fine-tune the GAN. The reservoir is constructed by resampling truth-level NuWro events according to flux-weighted target occupancies inferred from the MicroBooNE measurement. The visible bin structures reflect the experimental binning.

During base training, the real and generated samples are constructed with the same EνE_\nu spectrum, preventing the discriminator from learning to distinguish them from the neutrino-energy distribution alone. During fine-tuning, the generator instead samples EνE_\nu from the flux reweighted by the current σ\sigma-DNN prediction. Only the final layers of both the generator and discriminator are updated, while the preceding layers remain fixed. The σ\sigma-DNN is also partially frozen, with only its final nine Keras layers left trainable during fine-tuning. This preserves most of the representation learned during base training while allowing the generated kinematics to adapt to the data-informed reservoir.

During fine-tuning, the discriminator is updated on every reservoir batch. Every fifth discriminator update is followed by an adversarial update of the generator and then by the χc2\chi_c^2 update of the generator and the σ\sigma-DNN.

Role of the reservoir

An important subtlety concerns what the reservoir represents. Its target occupancies are derived from the smeared measurement and the MicroBooNE flux, whereas the reservoir itself consists of truth-level NuWro events. It therefore cannot be interpreted as a truth-level estimate of the measurement: the smearing matrix AcA_c is not inverted at any point, and such an inversion would in any case be ill-posed given the regularized unfolding.

In the adversarial sub-step, the reservoir provides the real-event sample for the update of the discriminator and, through the discriminator, of the generator. Unlike the χc2\chi_c^2 objective, the adversarial loss does not compare binned distributions; rather, it constrains the generated event distribution implicitly through the discriminator. In this way, the reservoir regularizes the generator toward a data-informed but physically valid, NuWro-like manifold of events and helps prevent mode collapse.

The quantitative comparison with the data is carried out in the second sub-step: the generated events are histogrammed, normalized using the σ\sigma-DNN prediction, forward-smeared by AcA_c, and compared with the original signed measurement through the χc2\chi_c^2 objective. Because this comparison is performed directly in the measurement space, it constrains both the shape and the normalization.

The two sub-steps thus play complementary roles. Without the first sub-step, the χc2\chi_c^2 term alone reproduces the smeared, flux-averaged distributions but distorts the QE and Δ\Delta-resonance structure (Figure 3 of the main text), which the smearing AcA_c and the coarse binning render invisible to χc2\chi_c^2. The reservoir supplies this event-level differential structure, while the χc2\chi_c^2 term performs the actual fit to the MicroBooNE data.

Second sub-step: differentiable matching to the differential cross section

In the second sub-step the generator GG and the σ\sigma-DNN are updated together so that the generated, cross-section-weighted, forward-smeared distribution matches the MicroBooNE differential measurement. The central difficulty is that this matching requires histogramming the generated events, which is not differentiable; we resolve it with a smooth histogram approximation described below.

Flux subdivision and the flux-averaged cross section

The measurement bins ω\omega are generally wider than, and not aligned with, the flux bins Φ\Phi. We therefore subdivide each Φ\Phi bin into finer sub-bins Φ′\Phi' of width ΔEν′\Delta E'_\nu, preserving the flux normalization within each parent bin, so that every measurement region ω\omega contains an integer number of Φ′\Phi' bins (Figure S5). For a region ω\omega the flux-averaged cross section can then be obtained in two equivalent ways,

Figure S5. Left: A representation of a flux Φ\Phi bin of width ΔEν\Delta E_\nu divided into sub-bins of flux Φ′\Phi' of width ΔEν′\Delta E'_\nu. Right: Measurement bins of width ΔEν,ω′′\Delta E''_{\nu,\omega} that may contain several Φ\Phi bins or fractions, but always entire Φ′\Phi' bins.

⟨σ^ω⟩=∑i∈ωσ^iΦi′∑i∈ωΦi′⏟prediction=⟨σ^Ω⟩nω/N∑i∈ωΦi′⏟MC,(18)\langle\hat{\sigma}_{\omega} \rangle= \underbrace{\frac{\sum_{i\in\omega} \hat{\sigma}_i\Phi'_i}{\sum_{i\in\omega}\Phi'_i}}_{\text{prediction}} = \langle\hat{\sigma}_{\Omega}\rangle \underbrace{\frac{n_\omega/N}{\sum_{i\in\omega}\Phi'_i}}_{\text{MC}}, \tag*{(18)}

the first using only the σ\sigma-DNN output σ^i\hat{\sigma}_i, the second relying on the sampled events (nωn_\omega of NN falling in ω\omega). The two expressions are analytically equivalent. In the optimization, we use the “prediction” expression based directly on the σ\sigma-DNN output, while the “MC” expression makes explicit the equivalent relation between the flux-averaged cross section and the fraction nω/Nn_\omega/N of generated events falling in the region ω\omega.

Differentiable histogram

The generator output (Eμ′,θ′)(E'_\mu,\theta') is first mapped to the physical variables (pμ,cos⁡θ)(p_\mu,\cos\theta). To histogram these events differentiably, we replace the hard assignment of an event to a bin by a smooth one built from sigmoids [41], S(x)=1/(1+e−x)S(x)=1/(1+e^{-x}). For a bin with edges (xi,xi+1)(x_i,x_{i+1}) and width Δx\Delta x, the contribution of the nn-th event, with value xnx_n, is the difference of two sigmoids,

hi(xn)=S(αxn−xiΔx)−S(αxn−xi+1Δx),(19)h_i(x_n)=S\left(\alpha\frac{x_n-x_i}{\Delta x}\right)-S\left(\alpha\frac{x_n-x_{i+1}}{\Delta x}\right), \tag*{(19)}

where α\alpha ( = 100) controls the sharpness of the bin edges. The first sigmoid turns on once xnx_n passes the left edge, the second once it passes the right edge; their difference is ≈1\approx1 for an event inside the bin and ≈0\approx0 outside, recovering hard binning as α→∞\alpha\to\infty while remaining smooth and differentiable for finite α\alpha. The soft bin count is

n~i=∑n=1Nhi(xn),(20)\widetilde{n}_i = \sum_{n=1}^{N} h_i(x_n), \tag*{(20)}

and for the two-dimensional (pμ,cos⁡θ)(p_\mu,\cos\theta) distribution the joint bin is the product

hij(xn,yn)=hi(xn)hj(yn),n~ij=∑n=1Nhij(xn,yn).(21)h_{ij}(x_n,y_n) = h_i(x_n)h_j(y_n), \qquad\widetilde{n}_{ij} = \sum_{n=1}^{N} h_{ij}(x_n,y_n). \tag*{(21)}

Because n~ij\widetilde{n}_{ij} is a smooth function of the event positions, and these depend on the generator weights, the gradient of the loss propagates through the histogram back to GG; a hard histogram would give a vanishing gradient and no training signal. This is the standard soft-histogram construction [41]; its accuracy to exact binning at our chosen α\alpha is shown in Figure S6. The soft count is an estimator of the exact count; events near a bin edge are shared between neighbors, negligible at large α\alpha but validated in Figure S6.

Figure S6. Left: double differential cross section as a function of EνE_\nu from the base DNN models, using exact event counts to create histograms, and the result using the differentiable histogram approximation, with a batch of 4000 events. Right: same data as left, but smeared with the matrix AcA_c.

From histogram to smeared prediction

The soft histogram is reweighted by the flux-averaged cross section to form an estimate of the double-differential cross section in region ω\omega,

(d2σ^ωdx dy)ij=⟨σ^ω⟩n~ωijnωΔxiΔyj,(22)\left(\frac{d^2\widehat{\sigma}_\omega}{dx\,dy}\right)_{ij} = \langle\widehat{\sigma}_\omega\rangle\frac{\widetilde{n}_{\omega ij}}{n_\omega\Delta x_i \Delta y_j}, \tag*{(22)}

where nωn_\omega is the number of generated events with EνE_\nu in the energy interval ω\omega. Eq. (S-IV.12) is then forward-smeared by the additional smearing matrix, s^ω,ij=Acσ^ω,ij\widehat{s}_{\omega,ij} = A_c\widehat{\sigma}_{\omega,ij}, bringing the prediction into the same space as the published measurement. The smearing is applied to the prediction and never inverted on the data, so the comparison is well-posed.

Gradient flow and the role of the σ\sigma-DNN

The σ\sigma-DNN enters this sub-step in two distinct roles. As the reweighter that defines the sampling distribution of EνE_\nu, it is treated as a fixed input and does not contribute to the gradient. As the predictor of the flux-averaged cross section ⟨σ^ω⟩\langle\widehat{\sigma}_\omega\rangle that rescales the histogram [Eq. S-IV.12], it is differentiated, and this is the path through which its weights are updated. Thus the sampling of EνE_\nu carries no gradient, while the cross-section normalization does.

Loss

The generator and σ\sigma-DNN weights are updated by minimizing the full-covariance χ2\chi^2,

χc2=∑kl(sk−s^k)Σkl−1(sl−s^l),(23)\chi_{\mathrm{c}}^2 = \sum_{kl} (s_k-\hat{s}_k)\Sigma^{-1}_{kl}(s_l-\hat{s}_l), \tag*{(23)}

where sks_k is the measured smeared cross section, s^k\hat{s}_k the smeared prediction of Eq. S-IV.12, Σ\Sigma the measurement covariance, and k,lk,l index the flattened (ω,i,j)(\omega,i,j) bins. Additionally, as a complementary diagnostic, we also report the diagonal metric

χv2=∑ωij(sωij−s^ωij)2Var⁡ωij,(24)\chi_{\mathrm{v}}^2 = \sum_{\omega ij} \frac{(s_{\omega ij}-\hat{s}_{\omega ij})^2}{\operatorname{Var}_{\omega ij}}, \tag*{(24)}

where Var⁡ωij\operatorname{Var}_{\omega ij} is the variance of sωijs_{\omega ij}.

Training results

In Figure S7, we repeat Figure 2 with additional plots of (MC−Data)/Uncertainty to show the quality of our model’s predictions.

Figure S7. Base and tuned models compared with MicroBooNE two-dimensional differential cross-section data in EνE_\nu Bins [23] on the left, and the (MC−Data)/Unc ratio on the right. The black plot in the ratio figure shows Data−|Data| to highlight the negative cross-section values found in the experimental measurement, and the gray bands show the measurement uncertainties. Grey vertical dashed lines show the different EνE_\nu regions.

Comparison with out-of-training MicroBooNE measurements

In Table S1, we present the χ2\chi^2 values for comparison with MicroBooNE measurements [24] for σ/⟨Eν⟩\sigma/\langle E_\nu\rangle, dσ/dEμ\mathrm{d}\sigma/\mathrm{d}E_\mu and dσ/dEhad\mathrm{d}\sigma/\mathrm{d}E_{\mathrm{had}} cross sections. These numbers refer to Figure 4.

σ(Eν)\sigma(E_\nu)dσ/dEμ\mathrm{d}\sigma/\mathrm{d}E_\mudσ/dEhad\mathrm{d}\sigma/\mathrm{d}E_{\mathrm{had}}
χc2/n\chi_{\mathrm{c}}^2/nχv2/n\chi_{\mathrm{v}}^2/nχc2/n\chi_{\mathrm{c}}^2/nχv2/n\chi_{\mathrm{v}}^2/nχc2/n\chi_{\mathrm{c}}^2/nχv2/n\chi_{\mathrm{v}}^2/n
Base58.1/1054.2/1029.1/1169.1/1117.1/841.3/8
Tuned44.2/100.5/1012.7/116.4/114.4/86.0/8

Table S1. χc2\chi_{\mathrm{c}}^2 and χv2\chi_{\mathrm{v}}^2 over the number of bins of the base and tuned DNN models for σ(Eν)\sigma(E_\nu), dσ/dEμ\mathrm{d}\sigma/\mathrm{d}E_\mu, and dσ/dEhad\mathrm{d}\sigma/\mathrm{d}E_{\mathrm{had}}.

In Figure S8, we compare our models’ predictions for the d2σ/dEμdcos⁡θd^{2}\sigma/dE_{\mu}d\cos\theta and dσ/dcos⁡θd\sigma/d\cos\theta cross-section distributions. Table S2 gives the χc2\chi^{2}_{c} and χv2\chi^{2}_{v} values for the base and tuned models.

Figure S8. Base model, tuned model, and MicroBooNE data [25, 26] for d2σ/dEμdcos⁡θd^{2}\sigma/dE_{\mu}d\cos\theta (left) and dσ/dcos⁡θd\sigma/d\cos\theta (right).

d2σ/dEμdcos⁡θd^{2}\sigma/dE_{\mu}d\cos\theta χc2/n\chi^{2}_{c}/nd2σ/dEμdcos⁡θd^{2}\sigma/dE_{\mu}d\cos\theta χv2/n\chi^{2}_{v}/ndσ/dcos⁡θd\sigma/d\cos\theta χc2/n\chi^{2}_{c}/ndσ/dcos⁡θd\sigma/d\cos\theta χv2/n\chi^{2}_{v}/n
Base257.0/69554.7/69291.5/172991.7/17
Tuned109.2/6987.9/696.8/174.8/17

Table S2. Same as Table S1 but for d2σ/dEμdcos⁡θd^{2}\sigma/dE_{\mu}d\cos\theta and dσ/dcos⁡θd\sigma/d\cos\theta.

ArgoNeuT results for neutrinos in neutrino mode.

In Figure S9 we present both distributions of cross sections reported by ArgoNeuT [27, 28], namely dσ/dθd\sigma/d\theta and dσ/dpμd\sigma/dp_{\mu}. Table S3 compares the base and tuned models to the ArgoNeuT dσ/dθd\sigma/d\theta and dσ/dpμd\sigma/dp_{\mu} measurements.

Figure S9. Base model, tuned model, and ArgoNeuT data [27, 28]. Left: dσ/dθd\sigma/d\theta. Right: dσ/dpμd\sigma/dp_{\mu}. Results for neutrinos in the NuMI neutrino beam. The shaded areas on the right-side plot show the pμp_{\mu} range of the training MicroBooNE data.

χν2/n\chi^2_\nu/n
dσ/dθd\sigma/d\thetadσ/dpμd\sigma/dp_\mu
Base13.2/146.1/11
Tuned11.4/146.0/11

Table S3. Comparison of MicroBooNE-tuned DNN to ArgoNeuT measurement of muon neutrino νμ\nu_\mu on NuMI neutrino beam.

ArgoNeuT results for neutrinos in antineutrino mode

In this section, we compare predictions from our system with ArgoNeuT measurements for neutrinos in an antineutrino beam. Table S4 presents the χν2\chi^2_\nu values computed for fine-tuned and base models. In Figure S10 we show the distributions of dσ/dθd\sigma/d\theta and dσ/dpμd\sigma/dp_\mu cross sections and predictions of fine-tuned and base models.

χν2/n\chi^2_\nu/n
dσ/dθd\sigma/d\thetadσ/dpμd\sigma/dp_\mu
Base9.1/189.6/20
Tuned9.6/189.1/20

Table S4. Comparison of MicroBooNE-tuned DNN to ArgoNeuT measurement of muon neutrino νμ\nu_\mu on NuMI anti-neutrino νˉμ\bar{\nu}_\mu beam.

Figure S10. Base model, tuned model, and ArgoNeuT data [27, 28]. Left: dσ/dθd\sigma/d\theta. Right: dσ/dpμd\sigma/dp_\mu. Results for neutrinos on the NuMI antineutrino beam. The shaded areas on the right-side plots show the pμp_\mu range of the training MicroBooNE data.

References

References

  1. [11]Y. Hayato and I. Pickering, Eur. Phys. J. ST 230, 4469 (2021), arXiv:2106.15809 [hep-ph].DOI
  2. [12]J. Isaacson, W. I. Jay, A. Lovato, P. A. N. Machado, and N. Rocco, Phys. Rev. D 107, 033007 (2023), arXiv:2205.06378 [hep-ph].DOI
  3. [13]J. M. Campbell et al., SciPost Phys. 16, 130 (2024), arXiv:2203.11110 [hep-ph].DOI
  4. [14]S. Pina-Otey, V. Gaitan, F. Sánchez, and T. Lux, Phys. Rev. D 102, 013003 (2020), arXiv:2005.12719 [hep-ex].
  5. [15]M. El Baz, F. Sánchez, N. Jachowicz, K. Niewczas, A. K. Jha, and A. Nikolakopoulos, Phys. Rev. D 111, 113001 (2025), arXiv:2502.14452 [hep-ex].arxiv.org/abs/2502.14452
  6. [16]Y. Alanazi et al., Phys. Rev. D 106, 096002 (2022), arXiv:2008.03151 [hep-ph].arxiv.org/abs/2008.03151
  7. [17]J. L. Bonilla, K. M. Graczyk, A. M. Ankowski, R. D. Banerjee, B. E. Kowal, H. Prasad, and J. T. Sobczyk, Phys. Rev. D 112, 013007 (2025).
  8. [18]J. L. Bonilla, K. M. Graczyk, A. M. Ankowski, R. D. Banerjee, B. E. Kowal, H. Prasad, and J. T. Sobczyk, Phys. Rev. D 113, 053001 (2026), arXiv:2508.12987 [hep-ph].DOI
  9. [19]K. M. Graczyk, B. E. Kowal, A. M. Ankowski, R. D. Banerjee, J. L. Bonilla, H. Prasad, and J. T. Sobczyk, Phys. Rev. Lett. 135, 052502 (2025).
  10. [20]K. Abe et al. (T2K), Nature 580, 339 (2020), [Erratum: Nature 583, E16 (2020)], arXiv:1910.03887 [hep-ex].DOI
  11. [21]A. Andreassen, P. T. Komiske, E. M. Metodiev, B. Nachman, and J. Thaler, Phys. Rev. Lett. 124, 182001 (2020), arXiv:1911.09107 [hep-ph].DOI
  12. [22]P. Abratenko et al. (MicroBooNE), Phys. Rev. D 105, 072001 (2022), arXiv:2110.14028 [hep-ex].DOI
  13. [23]P. Abratenko et al. (MicroBooNE), Phys. Lett. B 870, 139939 (2025), arXiv:2307.06413 [hep-ex].DOI
  14. [24]P. Abratenko et al. (MicroBooNE), Phys. Rev. Lett. 128, 151801 (2022), arXiv:2110.14023 [hep-ex].DOI
  15. [25]P. Abratenko et al. (MicroBooNE), Phys. Rev. Lett. 133, 041801 (2024), arXiv:2402.19281 [hep-ex].DOI
  16. [26]P. Abratenko et al. (MicroBooNE), Phys. Rev. D 110, 013006 (2024), arXiv:2402.19216 [hep-ex].DOI
  17. [27]C. Anderson et al. (ArgoNeuT), Phys. Rev. Lett. 108, 161802 (2012), arXiv:1111.0103 [hep-ex].DOI
  18. [28]R. Acciarri et al. (ArgoNeuT), Phys. Rev. D 89, 112003 (2014), arXiv:1404.4809 [hep-ex].DOI
  19. [29]See Supplemental Material for additional details on the cross-section DNN, GAN architecture and training, fine-tuning procedure, differentiable histogramming, and comparisons with out-of-training MicroBooNE and ArgoNeuT measurements.
  20. [30]S. Abe et al., Phys. Rev. D 113, 012011 (2026), arXiv:2509.17945 [hep-ex].DOI
  21. [31]P. Adamson et al., Nucl. Instrum. Meth. A 806, 279 (2016), arXiv:1507.06690 [physics.acc-ph].DOI
  22. [32]J. L. Bonilla and K. M. Graczyk, Numugen, https://github.com/LuisBonillaR/NuMuGen (2026).
  23. [33]T. Dozat, in Proceedings of the 4th International Conference on Learning Representations, Workshop Track (San Juan, Puerto Rico, 2016) pp. 1–4.
  24. [34]F. Chollet et al., Keras, https://keras.io (2015).
  25. [35]J. L. Ba, J. R. Kiros, and G. E. Hinton, Layer normalization (2016), arXiv:1607.06450 [stat.ML].arxiv.org/abs/1607.06450
  26. [36]X. Glorot, A. Bordes, and Y. Bengio, in Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, Proceedings of Machine Learning Research, Vol. 15, edited by G. Gordon, D. Dunson, and M. Dudík (PMLR, Fort Lauderdale, FL, USA, 2011) pp. 315–323.
  27. [37]K. He, X. Zhang, S. Ren, and J. Sun, arXiv e-prints , arXiv:1502.01852 (2015), arXiv:1502.01852 [cs.CV].
  28. [38]K. He, X. Zhang, S. Ren, and J. Sun, CoRR abs/1512.03385 (2015), 1512.03385.
  29. [39]G. Huang, Z. Liu, and K. Q. Weinberger, CoRR abs/1608.06993 (2016), 1608.06993.
  30. [40]S. J. Reddi, S. Kale, and S. Kumar, CoRR abs/1904.09237 (2019), 1904.09237.
  31. [41]M. Avi-Aharon, A. Arbelle, and T. R. Raviv, Deephist: Differentiable joint and color histogram layers for image-to-image translation (2020), arXiv:2005.03995 [eess.IV].

Paper details

vlnc:W_7vxzc42f
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents