Graczyk-Web-Site

Krzysztof Graczyk Homepage

home research info publications teaching neural networks fractals

Popular review abut neural networks in physics

Krzysztof M. Graczyk, Maciej Matyka
Głębokie sieci neuronowe dla fizyki,
Chapter of: BadAI. Nauka w erze sztucznej inteligencji. Eds. D. Buraczewski, M. Biernacka, P. Rychlikowski, Wydawnictwo Uniwersytetu Wrocławskiego, Wrocław 2025, p 35-54, ISBN978-83-68619-87-4. <<<<<

Machine Learning for Simulation of Neutrino-Nucleus Interactions

Transfer Learning for Neutrino Scattering: Domain Adaptation with GANs,
Jose L. Bonilla, Krzysztof M. Graczyk, Artur M. Ankowski, Rwik Dharmapal Banerjee, Beata E. Kowal, Hemant Prasad, Jan T. Sobczyk,
Phys. Rev. D 113, 053001 (2026)
Abstract:
Transfer learning (TL) is used to extrapolate the physics information encoded in a generative adversarial network trained on synthetic neutrino-carbon inclusive scattering data to related processes such as neutrino-argon and antineutrino-carbon interactions. We investigate how much of the underlying lepton-nucleus dynamics is shared across different targets and processes. We also assess the effectiveness of TL when training data is obtained from a different neutrino-nucleus interaction model. Our results show that TL not only reproduces key features of lepton kinematics, including the quasielastic and ?-resonance peaks, but also significantly outperforms generative models trained from scratch. Using data sets of 10,000 and 100,000 events, we find that TL maintains high accuracy even with limited statistics. Our findings demonstrate that TL provides a well-motivated and efficient framework for modeling (anti)neutrino-nucleus interactions and for constructing next-generation neutrino-scattering event generators, particularly valuable when experimental data are sparse.

Re-optimization of a Deep Neural Network Model for Electron–Carbon Scattering Using New Experimental Data,
Beata E. Kowal, Krzysztof M. Graczyk, Artur M. Ankowski, Rwik Dharmapal Banerjee, Jose L. Bonilla, Hemant Prasad, Jan T. Sobczyk,
Phys.Rev.C112, 055504 (2025)
Abstract:

We present an updated deep neural network model for inclusive electron–carbon scattering. Using the bootstrap model from Ref [Phys.Rev.C 110 (2024) 2, 025501] as a prior, we incorporate recent experimental data, including old deep inelastic scattering measurements, to derive a re-optimized posterior model. We examine the impact of these new inputs on model predictions and associated uncertainties. Finally, we evaluate the resulting cross-section predictions in the kinematic range relevant to the HyperKamiokande, and DUNE experiments.

Electron-nucleus cross sections from transfer learning,
Krzysztof M. Graczyk, Beata E. Kowal, Artur M. Ankowski, Rwik Dharmapal Banerjee, Jose L. Bonilla, Hemant Prasad, Jan T. Sobczyk,
Phys. Rev. Lett. 135, 052502 (2025)
Abstract:

Transfer learning allows a deep neural network (DNN) trained on one type of data to be adapted for new problems with limited information. We propose to use the transfer learning technique in physics. The DNN learns the details of one process, and after fine-tuning, it makes predictions for related processes. We consider the DNNs, trained on inclusive electron-carbon scattering data, and show that after fine-tuning, they accurately predict cross sections for electron interactions with nuclear targets ranging from helium-3 to iron.

Generative adversarial neural networks for simulating neutrino interactions,
Jose L. Bonilla, Krzysztof M. Graczyk, Artur M. Ankowski, Rwik Dharmapal Banerjee, Beata E. Kowal, Hemant Prasad, Jan T. Sobczyk,
Phys.Rev.D112, 013007 (2025)
Abstract:

We propose a new approach to simulate neutrino scattering events as an alternative to the standard Monte Carlo generator approach. Generative adversarial neural network (GAN) models are developed to simulate charged current neutrino-carbon collisions in the few-GeV energy range. We consider a simplified framework to generate muon kinematic variables, specifically its energy and scattering angle. GAN models are trained on simulation data from nuwro Monte Carlo event generator. Two GAN models have been obtained: one simulating quasielastic neutrino-nucleus scatterings and another simulating all interactions at given neutrino energy. The models work for neutrino energy ranging from 300 MeV to 10 GeV. The performance of both models has been assessed using two statistical metrics. It is shown that both GAN models successfully reproduce the distribution of muon kinematics.

Empirical fits to inclusive electron-carbon scattering data obtained by deep-learning methods,
Beata E. Kowal, Krzysztof M. Graczyk, Artur M. Ankowski, Rwik Dharmapal Banerjee, Hemant Prasad, Jan T. Sobczyk,
Phys. Rev. C 110, 025501 (2024)
Abstract:

Employing the neural network framework, we obtain empirical fits to the electron-scattering cross sections for carbon over a broad kinematic region, extending from the quasielastic peak through resonance excitation to the onset of deep-inelastic scattering. We consider two different methods of obtaining such model-independent parametrizations and the corresponding uncertainties: based on the bootstrap approach and the Monte Carlo dropout approach. In our analysis, the ??2 defines the loss function, including point-to-point and normalization uncertainties for each independent set of measurements. Our statistical approaches lead to fits of comparable quality and similar uncertainties of the order of 7%. To test these models, we compare their predictions to test datasets excluded from the training process and theoretical predictions obtained within the spectral function approach. The predictions of both models agree with experimental measurements and theoretical calculations. We also perform a comparison to a dataset lying beyond the covered kinematic region, and find that the bootstrap approach shows better interpolation and extrapolation abilities than the one based on the dropout algorithm.

Physics Informed Neural Networks

Bayesian Reasoning for Physics Informed Neural Networks,
Krzysztof M. Graczyk, Kornel Witkowski,
Phys.Rev.E 113 (2026) 5, 055307.
Abstract:

We introduce an evidence-driven Bayesian formulation of physics-informed neural networks that enables automatic optimization of loss weights between PDE residuals, boundary conditions, and observational data. Unlike existing Bayesian PINN approaches based on sampling or variational inference, the proposed method uses a Laplace approximation to compute model evidence analytically, enabling efficient hyperparameter tuning and model comparison without posterior sampling. We demonstrate the method on the heat, wave, and Burgers’ equations, obtaining solutions in agreement with exact or reference results. In the Burgers’ equation example, we further show that the framework naturally integrates information from governing equations and noisy measurements, providing predictive uncertainties within a unified Bayesian setting.

Deep Learning in Porous Media

Deep learning for diffusion in porous media,
Krzysztof M. Graczyk, Dawid Strzelczyk and Maciej Matyka, Sci Rep 13, 9769 (2023)
Abstract:
We adopt convolutional neural networks (CNN) to predict the basic properties of the porous media. Two different media types are considered: one mimics the sand packings, and the other mimics the systems derived from the extracellular space of biological tissues. The Lattice Boltzmann Method is used to obtain the labeled data necessary for performing supervised learning. We distinguish two tasks. In the first, networks based on the analysis of the system’s geometry predict porosity and effective diffusion coefficient. In the second, networks reconstruct the concentration map. In the first task, we propose two types of CNN models: the C-Net and the encoder part of the U-Net. Both networks are modified by adding a self-normalization module [Graczyk et al. in Sci Rep 12, 10583 (2022)]. The models predict with reasonable accuracy but only within the data type, they are trained on. For instance, the model trained on sand packings-like samples overshoots or undershoots for biological-like samples. In the second task, we propose the usage of the U-Net architecture. It accurately reconstructs the concentration fields. In contrast to the first task, the network trained on one data type works well for the other. For instance, the model trained on sand packings-like samples works perfectly on biological-like samples. Eventually, for both types of the data, we fit exponents in the Archie’s law to find tortuosity that is used to describe the dependence of the effective diffusion on porosity.

Predicting Porosity, Permeability, and Tortuosity of Porous Media from Images by Deep Learning,
Krzysztof M. Graczyk and Maciej Matyka, Sci Rep 10, 21488 (2020)
Abstract:
Convolutional neural networks (CNN) are utilized to encode the relation between initial configurations of obstacles and three fundamental quantities in porous media: porosity (?), permeability (k), and tortuosity (T). The two-dimensional systems with obstacles are considered. The fluid flow through a porous medium is simulated with the lattice Boltzmann method. The analysis has been performed for the systems with ??(0.37,0.99) which covers five orders of magnitude a span for permeability k?(0.78,2.1×105) and tortuosity T?(1.03,2.74). It is shown that the CNNs can be used to predict the porosity, permeability, and tortuosity with good accuracy. With the usage of the CNN models, the relation between T and ? has been obtained and compared with the empirical estimate.

Uncertainties in Deep Learning Systems

MOZART GRANT (WCA ): Opracowanie metod oceny niepewności w klasyfikacji próbek mikrobiologicznych
(eng.: The estimate of uncertainties in the classification of microbiological samples)
Project from 01.10.2019 to 30.09.2020, work done at NeuroSys
Self-Normalized Density Map (SNDM) for Counting Microbiological Obejcts,
Krzysztof M. Graczyk, Jarosław Pawlowski, Sylwia Majchrowska, Tomasz Golan,
Sci Rep 12, 10583 (2022)
Abstract:

The statistical properties of the density map (DM) approach to counting microbiological objects on images are studied in detail. The DM is given by U 2-Net. Two statistical methods for deep neural networks are utilized: the bootstrap and the Monte Carlo (MC) dropout. The detailed analysis of the uncertainties for the DM predictions leads to a deeper understanding of the DM model’s deficiencies. Based on our investigation, we propose a self-normalization module in the network. The improved network model, called Self-Normalized Density Map (SNDM), can correct its output density map by itself to accurately predict the total number of objects in the image. The SNDM architecture outperforms the original model. Moreover, both statistical frameworks—bootstrap and MC dropout— have consistent statistical results for SNDM, which were not observed in the original model. The SNDM efficiency is comparable with the detector-base models, such as Faster and Cascade R-CNN detectors.

Electromagnetic and Weak Structure of the Nucleon Investigated with Bayesian Neural Networks

Nucleon axial form factor from a Bayesian neural-network analysis of neutrino-scattering data,
Luis Alvarez-Ruso, Krzysztof M. Graczyk, Eduardo Saul-Sala, Phys. Rev. C99, 025204 (2019)
Abstract:
The Bayesian approach for feed-forward neural networks has been applied to the extraction of the nucleon axial form factor from the neutrino-deuteron scattering data measured by the Argonne National Laboratory (ANL) bubble chamber experiment. This framework allows to perform a model-independent determination of the axial form factor from data.. When the low $0.05 < Q^2 < 0.10$ GeV$^2$ data is included in the analysis, the resulting axial radius disagrees with available determinations. Furthermore, a large sensitivity to the corrections from the deuteron structure is obtained. In turn, when the low-$Q^2$ region is not taken into account, with or without deuteron corrections, no significant deviations from the dipole ansatz have been observed. A more accurate determination of the nucleon axial form factor requires new precise measurements of neutrino-induced quasielastic scattering on hydrogen and deuterium.

Zemach moments of proton from Bayesian inference,
Krzysztof M. Graczyk and Cezary Juszczak, Phys. Rev. C91, 045205 (2015)
Abstract:
The first and the third Zemach moments are obtained, $\langle r \rangle_{(2)}= 1.1108\pm 0.0021 $ fm and $\langle r^3\rangle_{(2)}=2.889 \pm 0.008$ fm$^3$, from the Bayesian analysis of the elastic $ep$ scattering data. The quantitative discussion of the dependence of the results on the parametrization choice is presented and the corresponding systematic uncertainties are estimated -- about 0.6\% and 1.6\% for the first and the third Zemach moments respectively.

Applications of Neural Networks in Hadron Physics,
Krzysztof M. Graczyk and Cezary Juszczak, J.Phys. G42 (2015) 3, 034019
invited contribution to special issue of J.Phys. G: Nucl. Phys., "Enhancing the interaction between nuclear experiment and theory through information and statistics" (ISNET).
Abstract:
The Bayesian approach for the feed-forward neural networks is reviewed. Its potential for usage in hadron physics is discussed. As an example of the application the study of the the two-photon exchange effect is presented. We focus on the model comparison, the estimation of the systematic uncertainties due to the choice of the model, and the over-fitting. As an illustration the predictions of the cross sections ratio $d \sigma(e^+ p\to e^+ p)/d \sigma(e^- p\to e^- p)$ are given together with the estimate of the uncertainty due to the parametrization choice.

Proton Radius from Bayesian Inference,
Krzysztof M. Graczyk and Cezary Juszczak, Phys. Rev. C90, 054334 (2014).
Abstract:
The methods of Bayesian statistics are used to extract the value of the proton radius from the elastic $ep$ scattering data in a model independent way. To achieve that goal a large number of parametrizations (equivalent to neural network schemes) are considered and ranked by their conditional probability $P(\mathrm{parametrization}\,|\,\mathrm{data})$ instead of using the minimal error criterion. As a result the most probable proton radii values ($r_E^p=0.899\pm 0.003$ fm, $r_M^p=0.879\pm 0.007$ fm) are obtained and systematic error due to freedom in the choice of parametrization is estimated. Correcting the data for the two photon exchange effect leads to smaller difference between the extracted values of $r_E^p$ and $r_M^p$. The results disagree with recent muonic atom measurements.

Comparison of Neural Network and Hadronic Model Predictions of Two-Photon Exchange Effect,
Krzysztof M. Graczyk, Phys. Rev. C88, 065205 (2013)
Abstract:
Predictions for the two-photon exchange (TPE) correction to unpolarized $ep$ elastic cross section, obtained within two different approaches, are confronted and discussed in detail. In the first one the TPE correction is extracted from experimental data by applying the Bayesian neural network (BNN) statistical framework. In the other the TPE is given by box diagrams, with the nucleon and the $P_{33}$ resonance as the hadronic intermediate states. Two different form factor parametrizations for both the proton and the $P_{33}$ resonance are taken into consideration. Proton form factors are obtained from the global fit of the full model (with the TPE correction) to the unpolarized cross section data. Predictions of both methods agree well in the intermediate $Q^2$ range, $(1,3)$ GeV$^2$. Above $Q^2=3$ GeV$^2$ the agreement is on $2\sigma$ level. Below $Q^2=1$ GeV$^2$ the consistency between both approaches is broken. The values of the proton radius extracted within both models are given. In both cases predictions for VEPP-3 experiment have been obtained and confronted with the preliminary experimental results.

Two-Photon Exchange Effect Studied with Neural Networks,
Krzysztof M. Graczyk, Phys. Rev. C84, 034314 (2011)
Abstract:
An approach to the extraction of the two-photon exchange (TPE) correction from elastic ep scattering data is presented. The cross-section, polarization transfer (PT), and charge asymmetry data are considered. It is assumed that the TPE correction to the PT data is negligible. The form factors and TPE correcting term are given by one multidimensional function approximated by the feedforward neural network (NN). To find a model-independent approximation, the Bayesian framework for the NNs is adapted. A large number of different parametrizations is considered. The most optimal model is indicated by the Bayesian algorithm. The obtained fit of the TPE correction behaves linearly in ? but it has a nontrivial Q2 dependence. A strong dependence of the TPE fit on the choice of parametrization is observed.

The analytical form of the fits fit.pdf and the covariance matrix (order of parameters the same as in fit.pdf )
Analysis done with:
GraNet - the feedworward neural network C++ library (will be avialable soon).

Neural Network Parameterizations of Electromagnetic Nucleon Form Factors,
Krzysztof M. Graczyk, Piotr Płoński, Robert Sulej, JHEP (2010) 053
Abstract:
The electromagnetic nucleon form-factors data are studied with artificial feed forward neural networks. As a result the unbiased model-independent form-factor parametrizations are evaluated together with uncertainties. The Bayesian approach for the neural networks is adapted for chi2 error-like function and applied to the data analysis. The sequence of the feed forward neural networks with one hidden layer of units is considered. The given neural network represents a particular form-factor parametrization. The so-called evidence (the measure of how much the data favor given form factor model) is computed with the Bayesian framework and it is used to determine the best form factor parametrization.

Analysis is done with:
GraNet - the feedworward neural network C++ library (will be avialable soon).

NetMaker (written in C#) by Robert Sulej and Piotr Płoński.