Search

Generative Adversarial Networks (GANs) are a class of neural networks that have been widely used in the field of image-to-image translation. In this paper, we propose a streamlined image-to-image translation network with a simpler architecture compared to existing models. We investigate the relationship between GANs and autoencoders and provide an explanation for the efficacy of employing only the GAN component for tasks involving image translation. We show that adversarial for GAN models yields results comparable to those of existing methods without additional complex loss penalties. Subsequently, we elucidate the rationale behind this phenomenon. We also incorporate experimental results to demonstrate the validity of our findings.

We revisit $B\to K^*\pi$ and $B\to K\rho$ decay processes using the modified perturbative QCD approach, which is developed based on the perturbative QCD approach with a few improvements. A critical infrared cutoff scale $\mu_c$ is introduced to separate hard and soft contributions in the decay process. Hard contributions with scale $\mu$ larger than $\mu_c$ are calculated with perturbative QCD, and soft contributions with $\mu<\mu_c$ are described in terms of soft transition and production form factors. In addition, color-octet contributions are considered phenomenologically. The branching fractions and $CP$ violations of $B\to K^*\pi$, $K\rho$ decays are calculated. With appropriate parameter inputs, the theoretical results can be in good agreement with experimental data., Comment: 16 pages, 1 figure

We study the decay processes of $B\rightarrow \rho\eta$, $\rho\eta^{\prime}$ in perturbative QCD approach with a few improvements incorproated in it, where the contributions with large momentum transfer are calculated perturbatively, and the contributions with lower energy scale are treated by introuducing soft transition form factors. In addition, color-octet conributions are introduced which are essentially long-distance contributions. With resonanble input parameters selected, we find that the theoretical results of the branching ratios and $CP$ violations are all consistent with experimental data. We also predict some unmeasured quantities, which can be tested by experiment in the future., Comment: 19 pages, 4 figures

Sleep is crucial for human health, and EEG signals play a significant role in sleep research. Due to the high-dimensional nature of EEG signal data sequences, data visualization and clustering of different sleep stages have been challenges. To address these issues, we propose a two-stage hierarchical and explainable feature selection framework by incorporating a feature selection algorithm to improve the performance of dimensionality reduction. Inspired by topological data analysis, which can analyze the structure of high-dimensional data, we extract topological features from the EEG signals to compensate for the structural information loss that happens in traditional spectro-temporal data analysis. Supported by the topological visualization of the data from different sleep stages and the classification results, the proposed features are proven to be effective supplements to traditional features. Finally, we compare the performances of three dimensionality reduction algorithms: Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP). Among them, t-SNE achieved the highest accuracy of 79.8%, but considering the overall performance in terms of computational resources and metrics, UMAP is the optimal choice.

Traditional dietary assessment methods heavily rely on self-reporting, which is time-consuming and prone to bias. Recent advancements in Artificial Intelligence (AI) have revealed new possibilities for dietary assessment, particularly through analysis of food images. Recognizing foods and estimating food volumes from images are known as the key procedures for automatic dietary assessment. However, both procedures required large amounts of training images labeled with food names and volumes, which are currently unavailable. Alternatively, recent studies have indicated that training images can be artificially generated using Generative Adversarial Networks (GANs). Nonetheless, convenient generation of large amounts of food images with known volumes remain a challenge with the existing techniques. In this work, we present a simple GAN-based neural network architecture for conditional food image generation. The shapes of the food and container in the generated images closely resemble those in the reference input image. Our experiments demonstrate the realism of the generated images and shape-preserving capabilities of the proposed framework.

We study the nonleptonic decays of $B\to PP$ in the modified perturbative QCD approach, where $P$ stands for pseudoscalar mesons. Transverse momenta of partons and the Sudakov factor are included, which help to suppress the contributions of soft interactions. The wave function of the $B$ meson obtained from the relativistic potential model is used, and then the contributions in the infrared region can not be suppressed completely. So a soft cutoff scale and soft form factors are introduced. The contributions with the scale higher than the soft cutoff scale are calculated with perturbative QCD, while the contributions lower than the cutoff scale are replaced by the soft form factors. To explain experimental data, we find that contributions of color-octet operators for the quark-antiquarks in the mesons in the final state need to be considered. The contributions of the color-octet operators are parametrized by a few parameters with the help of SU(3) flavor symmetry and symmetry breaking. These parameters for color-octet contributions are universal for all the nonleptonic decay modes of the $B$ meson, where the mesons in the final state belong to the same flavor SU(3) nonet. Both the branching ratios and $CP$ violations are studied. We find that the theoretical calculation can well explain the experimental data of $B$ factories.

The application of transfer learning, leveraging knowledge from source domains to enhance model performance in a target domain, has significantly grown, supporting diverse real-world applications. Its success often relies on shared knowledge between domains, typically required in these methodologies. Commonly, methods assume identical feature and label spaces in both domains, known as homogeneous transfer learning. However, this is often impractical as source and target domains usually differ in these spaces, making precise data matching challenging and costly. Consequently, heterogeneous transfer learning (HTL), which addresses these disparities, has become a vital strategy in various tasks. In this paper, we offer an extensive review of over 60 HTL methods, covering both data-based and model-based approaches. We describe the key assumptions and algorithms of these methods and systematically categorize them into instance-based, feature representation-based, parameter regularization, and parameter tuning techniques. Additionally, we explore applications in natural language processing, computer vision, multimodal learning, and biomedicine, aiming to deepen understanding and stimulate further research in these areas. Our paper includes recent advancements in HTL, such as the introduction of transformer-based models and multimodal learning techniques, ensuring the review captures the latest developments in the field. We identify key limitations in current HTL studies and offer systematic guidance for future research, highlighting areas needing further exploration and suggesting potential directions for advancing the field., Comment: 23 pages, 7 figures

With the increasing demands in communication, Wi-Fi technology is advancing towards its next generation. As high-need applications like Virtual Reality (VR) and Augmented Reality (AR) emerge, the role of millimeter-wave (mmWave) technology becomes critical. This paper explores Wi-Fi 8's potential features, especially its integration of mmWave technology. We address the challenges of implementing mmWave under current protocols and examine the compatibility of new features with mmWave. Our study includes system-level simulations, upclocking the 802.11ac PPDU to 60 GHz, and considers hardware limitations. The results demonstrate significant performance improvements with mmWave in Wi-Fi 8, indicating its feasibility for high-demand wireless scenarios.

We calculate the branching ratio and $CP$ violations for $B\to K\pi$ decays in a modified perturbative QCD approach based on $k_{T}$ factorization. The resummation effect of the transverse momentum regulates the endpoint singularity. Using the $B$ meson wave function that is obtained in the relativistic potential model, soft contribution cannot be suppressed effectively by Sudakov factor. Soft scale cutoff and soft $BK$, $B\pi$ and $K\pi$ form factors have to be introduced. The most important next-to-leading-order contributions from the vertex corrections, the quark loops, and the magnetic penguins are also considered. In addition, the contribution of the color-octet hadronic matrix element is included which is essentially of long-distance dynamics. Our predictions for branching ratios and $CP$ violations are in good agreement with the experimental data. Especially the theoretical result of dramatic difference between the $CP$ violations of $B^+\to K^+\pi^0$ and $B^0\to K^+\pi^-$ is consistent with experimental measurement, therefore the $K\pi$ puzzle in $B$ decays can be resolved in our way of the modified perturbative QCD approach., Comment: 19 pages, 2 figures, version to appear in Phys. Rev. D

We study $B\to \pi\pi$ decay with a modified perturbative QCD approach. The branching ratios and $CP$ violation are calculated with the transverse momenta of partons considered. Sudakov factor associated with each meson is included to suppress soft contribution in QCD. With the wave function of $B$ meson obtained in QCD-inspired relativistic potential model being used, the suppression of Sudakov factor to the soft contribution is not effective enough. Soft scale cutoff and soft form factors of $B\pi$ transition and $\pi\pi$ production have to be introduced. The main next-to-leading-order contributions of vertex correction, the quark-loop and magnetic penguin are included. To solve the long-standing puzzle in $B\to\pi\pi$ decay, that is the theoretical prediction of the branching ratio of $B^0\to \pi^0\pi^0$ being seriously smaller than experimental data, color-octet matrix element which is of long-distance dynamics is introduced. With parameters taken with reasonable values, all the branching ratios and $CP$ violation are well consistent with experimental data., Comment: 13 pages, 1 figure, version to appear in Phys. Rev. D

Sampling is ubiquitous in machine learning methodologies. Due to the growth of large datasets and model complexity, we want to learn and adapt the sampling process while training a representation. Towards achieving this grand goal, a variety of sampling techniques have been proposed. However, most of them either use a fixed sampling scheme or adjust the sampling scheme based on simple heuristics. They cannot choose the best sample for model training in different stages. Inspired by "Think, Fast and Slow" (System 1 and System 2) in cognitive science, we propose a reward-guided sampling strategy called Adaptive Sample with Reward (ASR) to tackle this challenge. To the best of our knowledge, this is the first work utilizing reinforcement learning (RL) to address the sampling problem in representation learning. Our approach optimally adjusts the sampling process to achieve optimal performance. We explore geographical relationships among samples by distance-based sampling to maximize overall cumulative reward. We apply ASR to the long-standing sampling problems in similarity-based loss functions. Empirical results in information retrieval and clustering demonstrate ASR's superb performance across different datasets. We also discuss an engrossing phenomenon which we name as "ASR gravity well" in experiments.

We consider the renormalization-group evolution of the matrix element of $\langle 0| \bar{q}(z)_\beta [z,0]b(0)_\alpha| \bar{B}\rangle$, which can be used to define the distribution amplitudes for $B$ meson and widely applied in studies of $B$ meson decays. The contribution to the renormalization constant of the non-local operator $\bar{q}(z)_\beta [z,0]b(0)_\alpha$ is considered up to one-loop order in QCD. Since the quark fields in this operator are not directly coupled fields, momentum can not flow freely through this non-local operator. Momentum involved in this operator can be treated stringently in coordinate space. We find that the ultraviolet divergences regulated by dimensional parameter $\epsilon$ cancel with each other, and the evolution effect vanishes. The matrix element $\langle 0| \bar{q}(z)_\beta [z,0]b(0)_\alpha| \bar{B}\rangle$ escapes from the renormalization-group evolution. We then apply the matrix element in calculating $B\to\pi$ transition form factor, where the matrix element is obtained by using the $B$ meson wave function calculated in QCD-inspired potential model. By comparing with experimental data for the semileptonic decay of $B\to \pi \ell\nu$ and light-cone sum rule calculation, we analyse the perturbative and non-perturbative contributions to $B\to\pi$ transition form factor in the frame work of perturbative QCD approach. We find that the effectiveness of the suppression of Sudakov factor to soft contribution depends on the end-point behavior of $B$ meson wave function, and with the $B$-meson wave function used in this work, soft contribution can not be well suppressed. The hard contribution to the $B\to\pi$ transition form factor is about 59\%, and soft contribution can be as large as 41\% in the naive calculation. To make the perturbative calculation reliable, a soft momentum cutoff in the calculation and soft form factor have to be introduced., Comment: 11 pages, 2 figure, version to appear in Nucl. Phys. B

Although deep learning based methods have achieved great success in many computer vision tasks, their performance relies on a large number of densely annotated samples that are typically difficult to obtain. In this paper, we focus on the problem of learning representation from unlabeled data for semantic segmentation. Inspired by two patch-based methods, we develop a novel self-supervised learning framework by formulating the Jigsaw Puzzle problem as a patch-wise classification process and solving it with a fully convolutional network. By learning to solve a Jigsaw Puzzle problem with 25 patches and transferring the learned features to semantic segmentation task on Cityscapes dataset, we achieve a 5.8 percentage point improvement over the baseline model that initialized from random values. Moreover, experiments show that our self-supervised learning method can be applied to different datasets and models. In particular, we achieved competitive performance with the state-of-the-art methods on the PASCAL VOC2012 dataset using significant fewer training images.

Based on the mechanism of neutral meson mixing, we predict the branching fraction of the semileptonic decay of $D_s^+\to \pi^0 \ell^+ \nu_\ell$ ($\ell=e$, $\mu$) using recently measured branching fraction of $D_s^+\to \eta e^+\nu_e$ by BESIII experiment . We also give a formula that can describe the neutral meson mixing of $\pi^0$, $\eta$, $\eta'$ and a pseudoscalar gluonium $G$ in a unified way. The predicted branching fraction of $D_s^+\to\pi^0 e^+\nu_e$ decay is $\mathcal{B}(D_s^+\to\pi^0 e^+\nu_e)=(2.65\pm 0.38)\times 10^{-5}$. It is important to search for the decay $D_s^+\to \pi^0 \ell^+ \nu_\ell$ in experiment, in order to understand the mechanism of $\pi^0$ production in $D_s^+\to\pi^0$ transition in semileptonic decay. We also estimate the branching fraction of $D_s^+\to \pi^0\tau^+\nu_\tau$, which is the only kinematically allowed semi-tauonic decay mode of the charmed meson, since the mass value of $D_s^+$ meson is just slightly above the threshold of $\pi^0\tau^+$ generation in the semi-tauonic decay., Comment: 6 pages, 1 figure

Background: Elderly patients with MODS have high risk of death and poor prognosis. The performance of current scoring systems assessing the severity of MODS and its mortality remains unsatisfactory. This study aims to develop an interpretable and generalizable model for early mortality prediction in elderly patients with MODS. Methods: The MIMIC-III, eICU-CRD and PLAGH-S databases were employed for model generation and evaluation. We used the eXtreme Gradient Boosting model with the SHapley Additive exPlanations method to conduct early and interpretable predictions of patients' hospital outcome. Three types of data source combinations and five typical evaluation indexes were adopted to develop a generalizable model. Findings: The interpretable model, with optimal performance developed by using MIMIC-III and eICU-CRD datasets, was separately validated in MIMIC-III, eICU-CRD and PLAGH-S datasets (no overlapping with training set). The performances of the model in predicting hospital mortality as validated by the three datasets were: AUC of 0.858, sensitivity of 0.834 and specificity of 0.705; AUC of 0.849, sensitivity of 0.763 and specificity of 0.784; and AUC of 0.838, sensitivity of 0.882 and specificity of 0.691, respectively. Comparisons of AUC between this model and baseline models with MIMIC-III dataset validation showed superior performances of this model; In addition, comparisons in AUC between this model and commonly used clinical scores showed significantly better performance of this model. Interpretation: The interpretable machine learning model developed in this study using fused datasets with large sample sizes was robust and generalizable. This model outperformed the baseline models and several clinical scores for early prediction of mortality in elderly ICU patients. The interpretative nature of this model provided clinicians with the ranking of mortality risk features., Comment: 33 pages, 14 figures, 14 tables, article, Co-author: Xiaoli Liu and Pan Hu, Co-correspondence: Feihu Zhou and Zhengbo Zhang

In this work we study the semi-leptonic decay of $\bar{B}_s^0\to \phi l^+ l^-$($l=e, \mu, \tau$) with QCD sum rule method. We calculate the $\bar{B}_s^0\to \phi$ translation form factors relevant to this semi-leptonic decay, then the branching ratios of $\bar{B}_s^0\to \phi l^+ l^-$($l=e, \mu, \tau$) decays are calculated with the form factors obtained here. Our result for the branching ratio of $\bar{B}_s^0\to \phi\mu^+ \mu^-$ agrees very well with the recent experimental data. For the unmeasured decay modes such as $\bar{B}_s^0\to \phi e^+ e^-$ and $\bar{B}_s^0\to \phi\tau^+ \tau^-$, we give theoretical predictions., Comment: 25 pages, 4 figures, more content added

We calculate $\bar{B}_s^0\to \phi$ translation form factors $V$, $A_0$, $A_1$, $A_2$ based on QCD sum rule and study the nonleptonic two-body decay of $\bar{B}_s^0\to J/\psi \phi$ with the form factors obtained. We calculate the time-integrated branching ratio of $\bar{B}_s^0\to J/\psi \phi$ decay. The results for both the total branching ratio and the cases for the final vectors in longitudinal and transverse polarizations are consistent with experimental data., Comment: 32 pages, 8 figures, more content added

The semileptonic decays and two-body nonleptonic decays of light baryon octet ($T_8$) and decuplet ($T_{10}$) consisting of light $u,d,s$ quarks are studied with the SU(3) flavor symmetry in this work. We obtain the amplitude relations between different decay modes by the SU(3) irreducible representation approach, and then predict relevant branching ratios by present experimental data within $1 \sigma$ error. We find that the predictions for all branching ratios except $\mathcal{B}(\Xi\rightarrow \Lambda^0\pi)$ and $\mathcal{B}(\Xi^*\rightarrow \Xi\pi)$ are in good agreement with present experimental data, that implies the neglected $C_+$ terms or SU(3) breaking effects might contribute at the order of a few percent in $\Xi\rightarrow \Lambda^0\pi$ and $\Xi^*\rightarrow \Xi\pi$ weak decays. We predict that $\mathcal{B}(\Xi^{-}\rightarrow \Sigma^0\mu^-\bar{\nu}_\mu)=(1.13\pm0.08)\times10^{-6}$, $\mathcal{B}(\Xi^{-}\rightarrow\Lambda^0\mu^-\bar{\nu}_\mu)=(1.58\pm0.04)\times10^{-4}$, $\mathcal{B}(\Omega^-\rightarrow\Xi^0\mu^-\bar{\nu}_\mu)=(3.7\pm1.8)\times10^{-3}$, $\mathcal{B}(\Sigma^-\rightarrow \Sigma^0e^-\bar{\nu}_e)=(1.35\pm0.28)\times10^{-10}$, $\mathcal{B}(\Xi^-\rightarrow \Xi^0e^-\bar{\nu}_e)=(4.2\pm2.4)\times10^{-10}$. We also study $T_{10}\to T_8 P_8$ weak, electromagnetic or strong decays. Some of these decay modes could be observed by the BESIII, LHCb and other experiments in the near future. Due to the very small life times of $\Sigma^0$, $\Xi^{*0,-}$, $\Sigma^{*0,-}$ and $\Delta^{0,-}$, the branching ratios of these baryon weak decays are only at the order of $\mathcal{O}(10^{-20}-10^{-13}$), which are too small to be reached by current experiments. Furthermore, the longitudinal branching ratios of $T_{8A} \to T_{8B} \ell^- \bar{\nu}_\ell~(\ell=\mu,e)$ decays are also given., Comment: 23 pages, 4 figures

We study the electromagnetic Dalitz (EM) decays, $\psi \to \eta_{c} l^{+} l^{-}$ and $\Upsilon \to \eta_{b} l^{+} l^{-}$ ($l = e$ or $\mu$), in which the lepton pair comes from the virtual photon emitted by the M1 transition from $c\bar{c}$ ($b\bar{b}$) spin triplet state to the spin singlet state. We estimate the partial width of $\psi(\Upsilon) \to \eta_{c}(\eta_{b}) l^{+} l^{-}$, based on the simple pole approximation. Besides, based on different QCD models, the partial width of $\psi(\Upsilon) \to \eta_{c}(\eta_{b}) \gamma$ is determined.

The rapid development of autonomous driving in recent years presents lots of challenges for scene understanding. As an essential step towards scene understanding, semantic segmentation thus received lots of attention in past few years. Although deep learning based state-of-the-arts have achieved great success in improving the segmentation accuracy, most of them suffer from an inefficiency problem and can hardly applied to practical applications. In this paper, we systematically analyze the computation cost of Convolutional Neural Network(CNN) and found that the inefficiency of CNN is mainly caused by its wide structure rather than the deep structure. In addition, the success of pruning based model compression methods proved that there are many redundant channels in CNN. Thus, we designed a very narrow while deep backbone network to improve the efficiency of semantic segmentation. By casting our network to FCN32 segmentation architecture, the basic structure of most segmentation methods, we achieved 60.6\% mIoU on Cityscape val dataset with 54 frame per seconds(FPS) on $1024\times2048$ inputs, which already outperforms one of the earliest real time deep learning based segmentation methods: ENet., Comment: 6pages