Search

Dataset bias is a significant problem in training fair classifiers. When attributes unrelated to classification exhibit strong biases towards certain classes, classifiers trained on such dataset may overfit to these bias attributes, substantially reducing the accuracy for minority groups. Mitigation techniques can be categorized according to the availability of bias information (\ie, prior knowledge). Although scenarios with unknown biases are better suited for real-world settings, previous work in this field often suffers from a lack of interpretability regarding biases and lower performance. In this study, we propose a framework to identify potential biases as keywords without prior knowledge based on the partial occurrence in the captions. We further propose two debiasing methods: (a) handing over to an existing debiasing approach which requires prior knowledge by assigning pseudo-labels, and (b) employing data augmentation via text-to-image generative models, using acquired bias keywords as prompts. Despite its simplicity, experimental results show that our framework not only outperforms existing methods without prior knowledge, but also is even comparable with a method that assumes prior knowledge.

Structure functions of the spin-1 deuteron will be investigated experimentally from the late 2020's at various facilities such as Thomas Jefferson National Accelerator Facility, Fermi National Accelerator Laboratory, nuclotron-based ion collider facility, and electron-ion colliders. We expect that a new high-energy spin-physics field could be created by these projects. In this paper, the current theoretical status is explained for the structure functions of spin-1 hadrons, especially on parton distribution functions, transverse-momentum dependent parton distributions, and fragmentation functions. Related multiparton distribution functions are also shown., Comment: 12 pages, 20 figures, European Physical Journal A, topical issue on the subject matter of Tensor Spin Observables, in press

QCD gives rise to a rich spectrum of excited baryon states. Understanding their internal structure is important for many areas of nuclear physics, such as nuclear forces, dense matter, and neutrino-nucleus interactions. Generalized parton distributions (GPDs) are an established tool for characterizing the QCD structure of the ground-state nucleon. They are used to create 3D tomographic images of the quark/gluon structure and quantify the mechanical properties such as the distribution of mass, angular momentum and forces in the system. Transition GPDs extend these concepts to $N \rightarrow N^\ast$ transitions and can be used to characterize the 3D structure and mechanical properties of baryon resonances. They can be probed in high-momentum-transfer exclusive electroproduction processes with resonance transitions $e + N \rightarrow e' + M + N^\ast$, such as deeply-virtual Compton scattering ($M = \gamma$) or meson production ($M = \pi, K$, $etc.$), and in related photon/hadron-induced processes. This White Paper describes a research program aiming to explore baryon resonance structure with transition GPDs. This includes the properties and interpretation of the transition GPDs, theoretical methods for structures and processes, first experimental results from JLab 12 GeV, future measurements with existing and planned facilities (JLab detector and energy upgrades, COMPASS/AMBER, EIC, EicC, J-PARC, LHC ultraperihperal collisions), and the theoretical and experimental developments needed to realize this program.

This study presents a novel approach for generating holograms based on the neural radiance fields (NeRF) technique. Generating three-dimensional (3D) data is difficult in hologram computation. NeRF is a state-of-the-art technique for 3D light-field reconstruction from 2D images based on volume rendering. The NeRF can rapidly predict new-view images that do not include a training dataset. In this study, we constructed a rendering pipeline directly from a 3D light field generated from 2D images by NeRF for hologram generation using deep neural networks within a reasonable time. The pipeline comprises three main components: the NeRF, a depth predictor, and a hologram generator, all constructed using deep neural networks. The pipeline does not include any physical calculations. The predicted holograms of a 3D scene viewed from any direction were computed using the proposed pipeline. The simulation and experimental results are presented.

The definitions of the quark and gluon transversity generalized parton distributions (GPDs) in spin-3/2 particles are obtained in the light-cone gauge. It is found that they contain 16 independent components for each parton. Their even or odd property is found in terms of skewness variable, and the odd transversity GPDs vanish in the forward limit. There are 16 amplitudes with helicity flip of quark or gluon. We conclude that all the amplitudes, unpolarized, polarized, and transversity amplitudes, have a common factor $F(\zeta)$ carrying all the complex part. Here, the variable $\zeta$ is the helicity difference $\zeta=(\lambda' -\lambda)-(\mu' - \mu)$ for the amplitude $\mathcal{A}_{\lambda' \mu',\lambda\mu}$. This kinematical factor is associated with the transfer of the orbital angular momentum in the quark- and gluon-spin-3/2 particle scattering amplitudes. We also derive another main physical quantity, transversity distribution, from the transversity GPDs in the forward limit for the spin-3/2 particles., Comment: 22 pages, 2 figures

It is not fully understood why adversarial examples can deceive neural networks and transfer between different networks. To elucidate this, several studies have hypothesized that adversarial perturbations, while appearing as noises, contain class features. This is supported by empirical evidence showing that networks trained on mislabeled adversarial examples can still generalize well to correctly labeled test samples. However, a theoretical understanding of how perturbations include class features and contribute to generalization is limited. In this study, we provide a theoretical framework for understanding learning from perturbations using a one-hidden-layer network trained on mutually orthogonal samples. Our results highlight that various adversarial perturbations, even perturbations of a few pixels, contain sufficient class features for generalization. Moreover, we reveal that the decision boundary when learning from perturbations matches that from standard samples except for specific regions under mild conditions. The code is available at https://github.com/s-kumano/learning-from-adversarial-perturbations., Comment: ICLR24

Earthquakes are among the most immediate and deadly natural disasters that humans face. Accurately forecasting the extent of earthquake damage and assessing potential risks can be instrumental in saving numerous lives. In this study, we developed linear regression models capable of predicting seismic intensity distributions based on earthquake parameters: location, depth, and magnitude. Because it is completely data-driven, it can predict intensity distributions without geographical information. The dataset comprises seismic intensity data from earthquakes that occurred in the vicinity of Japan between 1997 and 2020, specifically containing 1,857 instances of earthquakes with a magnitude of 5.0 or greater, sourced from the Japan Meteorological Agency. We trained both regression and classification models and combined them to take advantage of both to create a hybrid model. The proposed model outperformed commonly used Ground Motion Prediction Equations (GMPEs) in terms of the correlation coefficient, F1 score, and MCC. Furthermore, the proposed model can predict even abnormal seismic intensity distributions, a task at conventional GMPEs often struggle.

Gravitational form factors of hadrons provide information on mass and pressure distributions in the hadrons, and they have been investigated by generalized parton distributions (GPDs). The GPDs also contain information on internal compositions of hadron spins. Understanding on the origin of hadron masses, pressures, and spins is an important topic not only in hadron physics but also as fundamental physics, and it should be clarified by the GPD studies. The spacelike and timelike GPDs have been measured at charged-lepton accelerator facilities, for example, by the virtual Compton scattering and the two-photon process, respectively. In this work, we show the $\pi^\pm$ and $\pi^0$ production cross sections in neutrino (antineutrino) reactions $\nu \,(\bar\nu) + N \to \ell + \pi + N'$ for measuring the GPDs of the nucleon by using the theoretical formalism of Pire, Szymanowski, and Wagner. The $\pi^\pm$-production cross sections are useful for determining gluon GPDs, whereas the $\pi^0$ production probes quark GPDs. In particular, we show the roles of the pion- and rho-pole terms, the contribution from each GPD term ($H$, $E$, $\tilde H$, $\tilde E$), the effects of the gluon GPDs $H_g$ and $E_g$, and the effects of the reaction energy on the neutrino cross sections. The $\pi^0$-production cross section is sensitive to the quark GPDs including the pion- and rho-pole GPD terms in the Efremov-Radyushkin-Brodsky-Lepage region. These cross sections could be measured at Fermilab in future. The neutrino GPD studies will play a complementary role to the projects of charged-lepton and hadron reactions for determining the accurate GPDs., Comment: 15 pages, 14 figures

We outline the physics opportunities provided by the Electron Ion Collider (EIC). These include the study of the parton structure of the nucleon and nuclei, the onset of gluon saturation, the production of jets and heavy flavor, hadron spectroscopy and tests of fundamental symmetries. We review the present status and future challenges in EIC theory that have to be addressed in order to realize this ambitious and impactful physics program, including how to engage a diverse and inclusive workforce. In order to address these many-fold challenges, we propose a coordinated effort involving theory groups with differing expertise is needed. We discuss the scientific goals and scope of such an EIC Theory Alliance., Comment: 44 pages, ReVTeX, White Paper on EIC Theory Alliance

This paper presents a comprehensive review of both the theory and experimental successes of Quantum Chromodynamics, starting with its emergence as a well defined theory in 1972-73 and following developments and results up to the present day. Topics include a review of the earliest theoretical and experimental foundations; the fundamental constants of QCD; an introductory discussion of lattice QCD, the only known method for obtaining exact predictions from QCD; methods for approximating QCD, with special focus on effective field theories; QCD under extreme conditions; measurements and predictions of meson and baryon states; a special discussion of the structure of the nucleon; techniques for study of QCD at high energy, including treatment of jets and showers; measurements at colliders; weak decays and quark mixing; and a section on the future, which discusses new experimental facilities or upgrades currently funded. The paper is intended to provide a broad background for Ph.D. students and postdocs starting their career. Some contributions include personal accounts of how the ideas or experiments were developed., Comment: Invited volume for the EJPC; 567 pages if text/figures and 4783 references occupying about 160 additional pages. arXiv abstract abridged, for the complete abstract please see the full text

Although Convolutional Neural Networks (CNNs) have high accuracy in image recognition, they are vulnerable to adversarial examples and out-of-distribution data, and the difference from human recognition has been pointed out. In order to improve the robustness against out-of-distribution data, we present a frequency-based data augmentation technique that replaces the frequency components with other images of the same class. When the training data are CIFAR10 and the out-of-distribution data are SVHN, the Area Under Receiver Operating Characteristic (AUROC) curve of the model trained with the proposed method increases from 89.22\% to 98.15\%, and further increased to 98.59\% when combined with another data augmentation method. Furthermore, we experimentally demonstrate that the robust model for out-of-distribution data uses a lot of high-frequency components of the image., Comment: ICIP 2022

The ground motion prediction equation is commonly used to predict the seismic intensity distribution. However, it is not easy to apply this method to seismic distributions affected by underground plate structures, which are commonly known as abnormal seismic distributions. This study proposes a hybrid of regression and classification approaches using neural networks. The proposed model treats the distributions as 2-dimensional data like an image. Our method can accurately predict seismic intensity distributions, even abnormal distributions., Comment: 2 pages, 2 figures, IEEE GCCE2022 accepted

Adversarial attacks have only focused on changing the predictions of the classifier, but their danger greatly depends on how the class is mistaken. For example, when an automatic driving system mistakes a Persian cat for a Siamese cat, it is hardly a problem. However, if it mistakes a cat for a 120km/h minimum speed sign, serious problems can arise. As a stepping stone to more threatening adversarial attacks, we consider the superclass adversarial attack, which causes misclassification of not only fine classes, but also superclasses. We conducted the first comprehensive analysis of superclass adversarial attacks (an existing and 19 new methods) in terms of accuracy, speed, and stability, and identified several strategies to achieve better performance. Although this study is aimed at superclass misclassification, the findings can be applied to other problem settings involving multiple classes, such as top-k and multi-label classification attacks., Comment: ICML Workshop 2022 on Adversarial Machine Learning Frontiers

The J-PARC is a hadron-accelerator facility to provide secondary beams of kaons, pions, neutrinos, muons, and the others together with the primary proton beam for investigating a wide range of science projects. High-energy hadron physics can be studied by using high-momentum beams of unseparated hadrons, which are essentially pions, and also primary protons. In this report, possible experiments are explained on color transparency and generalized parton distributions (GPDs). These projects are complementary to lepton-scattering experiments at JLab, COMPASS/AMBER, and future electron-ion colliders. Because of hadron-beam energies up to 30 GeV, the J-PARC is a unique facility to investigate the transition region from the hadron degrees of freedom to the quark-gluon one. It is suitable for finding mechanisms of the color transparency. Such color-transparency studies are also valuable for clarifying factorization of hadron-production processes in extracting the GPDs from actual measurements. These studies will lead to the understanding of basic high-energy hadron interactions in nuclear medium and to clarifications on the origins of hadron spins, masses, and internal pressure mechanisms., Comment: 8 pages, LaTeX, Proceedings of the Workshop on The Future of Color Transparency and Hadronization Studies at Jefferson Lab and Beyond, June 7-8, 2021, (Online) USA-eastern time (MSU/Orsay/FIU/Penn State)

Spacelike and timelike generalized parton distributions (GPDs) have been investigated in charged-lepton scattering and electron-positron collisions via deeply virtual Compton scattering and two-photon processes, respectively. Furthermore, we expect that hadron-accelerator-facility measurements will be performed in future. The GPDs will play a crucial role in clarifying the origins of hadron spins and masses in terms of quarks and gluons. It is also possible to probe internal pressure within hadrons for understanding their stability. Gravitational form factors of hadrons used to be considered as a purely academic subject because gravitational interactions are too weak to be measured in microscopic systems. However, due to the development of hadron-tomography field, it became possible to extract the gravitational form factors from the actual GPD measurements without relying on direct gravitational interactions. Neutrino reactions can also be used for GPD studies in future, for example, by using the Long-Baseline Neutrino Facility at Fermilab. The neutrino GPD measurements are valuable especially for finding the flavor dependence of the GPDs in a complementary way to the charged-lepton experiments. We give an overview of the GPDs and discuss possible neutrino GPD measurements using the single-pion production processes $\nu + N \to \ell^- + N' + \pi$ and $\bar\nu + N \to \ell^+ + N' + \pi$., Comment: 7 pages, LaTeX, Proceedings of the 22nd International Workshop on Neutrinos from Accelerators (NuFact2021), September 5-11, 2021, (Online/In person) Cagliari, Italy

There are two types of polarizations in spin-1 hadrons, and they are vector and tensor polarizations. The latter is a unique one since it does not exist in the spin-1/2 proton. The vector-polarized PDFs are the same for both the proton and spin-1 hadrons; therefore, we mainly investigate the unique PDFs in tensor-polarized hadrons. By using the operator product expansion, the twist-3 PDF $f_{LT}$ can be expressed by two terms in the same way with $g_T$ of the proton. The first term is determined by the twist-2 PDF $f_{1LL}$ (or $b_1$) which was measured by an experiment, and the second term is expressed by twist-3 quark-gluon distributions. If we neglect the higher-twist effects, $f_{LT}$ is simply given by $f_{1LL}$, and this relation is similar to the Wandzura-Wilczek relation of $g_T$. Furthermore, a new sum rule is also obtained for $f_{2LT}=2/3 f_{LT}- f_{1LL}$, which is analogous to the Burkhardt-Cottingham sum rule, in the tensor-polarized spin-1 hadrons. In future, these interesting relations could be studied at accelerator facilities, such as the Jefferson Laboratory, the Fermilab, the nuclotron-based ion collider facility in Russia, the proposed electron-ion colliders in US and China., Comment: 5 pages, 1 figure, Part of Proceedings, 24th International Spin Symposium, October 18-22, 2021, Matsue, Japan

Transverse-momentum-dependent parton distribution functions (TMDs) were investigated at the twists 3 and 4 for spin-1 hadrons in addition to the twist-2 ones. They were found by studying all the possible decomposition of a quark correlation function in a Lorentz-invariant way with the Hermiticity and parity invariance. The time-reversal invariance was not imposed for the TMDs due to an active role of gauge links; however, they were used for collinear parton distribution functions (PDFs) by integrating the TMDs over the transverse momentum. We found that 30 TMDs exist in the tensor-polarized spin-1 hadron at the twists 3 and 4, whereas there are 10 TMDs at the twist 2. We also showed that there are 3 collinear PDFs at the twists 3 and 4. The corresponding TMD fragmentation functions exist sat the twists 2, 3, and 4 simply by changing function names and variables. Since the time-reversal invariance is valid in the collinear PDFs, the integrals of time-reversal-odd TMDs over the transverse momentum should vanish. It leads to the sum rules $ \int d^2 k_T \, h_{1LT} (x, k_T^{\, 2}) = \int d^2 k_T g_{LT} (x, k_T^{\, 2}) = \int d^2 k_T h_{LL} (x, k_T^{\, 2}) = \int d^2 k_T h_{3LL}(x, k_T^{\, 2}) =0$ on the time-reversal odd TMDs at the twists 3 and 4., Comment: 5 pages, LaTeX, 24th International Spin Symposium, October 18-22, 2021, (In person/Online) Matsue, Japan

We explain a gluon transversity, transverse-momentum-dependent parton distribution functions (TMDs), and parton distribution functions (PDFs) for spin-1 hadrons. The gluon transversity exists in hadrons with spin more than or equal to one, and it does not exist in the spin-1/2 nucleons. Since there is no direct contribution from the nucleons, it is an appropriate quantity to probe an exotic component in the spin-1 deuteron beyond a simple bound system of the nucleons. We show how the gluon transversity can be measured at hadron accelerator facilities by the Drell-Yan process in addition to lepton-accelerator experiments. Next, possible TMDs are explained for the spin-1 hadrons at the twists 3 and 4 in addition to twist-2 ones by considering tensor polarizations. We found that 30 TMDs exist in the tensor-polarized spin-1 hadron at the twists 3 and 4 in addition to 10 TMDs at the twist 2. There are 3 collinear PDFs at the twists 3 and 4. We also indicate that the corresponding TMD fragmentation functions exist at the twists 3 and 4. Due to the time-reversal invariance in the collinear PDFs, there are new sum rules on the time-reversal odd TMDs. In addition, we obtained a useful twist-2 relation, a sum rule, and relations with multiparton distribution functions by using the operator product expansion and the equation of motion for quarks. These findings are valuable for experimental investigations on polarized deuteron structure functions in 2020's and 2030's at world accelerator facilities., Comment: 5 pages, LaTeX, 19th International Conference on Hadron Spectroscopy and Structure (Haron2021), July 26-31, 2021, (Online) Mexico City, Mexico

Structure functions of polarized spin-1 hadrons will be measured at various accelerator facilities in the near future. Recently, transverse-momentum-dependent and collinear parton distribution functions were theoretically proposed at twist 3 and twist 4 in addition to the twist-2 ones, so that full investigations became possible for structure functions of spin-1 hadrons in the same level with those of the spin-1/2 nucleons. Furthermore, twist-3 tensor-polarized multiparton distribution functions were also recently found for spin-1 hadrons. In this work, we show relations among the collinear parton- and multiparton-distribution functions for spin-1 hadrons by using equation of motion for quarks. These relations are valuable in constraining the distribution functions and learning about multiparton correlations in spin-1 hadrons., Comment: 5 pages