Search arXiv⌕ Search

arXiv subjects

Yong Zhao

Publications and source records attributed to Yong Zhao.

At least 127 records · Page 7Linked to original sources
arXiv2023

ONNXExplainer: an ONNX Based Generic Framework to Explain Neural Networks Using Shapley Values

Understanding why a neural network model makes certain decisions can be as important as the inference performance. Various methods have been proposed to help practitioners explain the prediction of a neural network model, of which Shapley values are most popular. SHAP package is a leading implementation of Shapley values to explain neural networks implemented in TensorFlow or PyTorch but lacks cross-platform support, one-shot deployment and is highly inefficient. To address these problems, we present the ONNXExplainer, which is a generic framework to explain neural networks using Shapley values in the ONNX ecosystem. In ONNXExplainer, we develop its own automatic differentiation and optimization approach, which not only enables One-Shot Deployment of neural networks inference and explanations, but also significantly improves the efficiency to compute explanation with less memory consumption. For fair comparison purposes, we also implement the same optimization in TensorFlow and PyTorch and measure its performance against the current state of the art open-source counterpart, SHAP. Extensive benchmarks demonstrate that the proposed optimization approach improves the explanation latency of VGG19, ResNet50, DenseNet201, and EfficientNetB0 by as much as 500%.

cs.LG↗
arXiv2023

Multiple-distribution-function finite-difference lattice Boltzmann method for incompressible Navier-Stokes equation

In this paper, a multiple-distribution-function finite-difference lattice Boltzmann method (MDF-FDLBM) is proposed for the convection-diffusion system based incompressible Navier-Stokes equations (NSEs). By Chapman Enskog analysis, the convection-diffusion system based incompressible NSEs can be recovered from MDF-FDLBM. Some quantities, including the velocity gradient, velocity divergence, strain rate tensor, shear stress and vorticity, can be computed locally by the first-order moment of the non-equilibrium distribution function. Through the von Neumann analysis, we conduct the stability analysis for the MDF-FDLBM and incompressible finite-difference lattice Boltzmann method (IFDLBM). It is found that the IFDLBM will be more stable than that of MDF-FDLBM with small kinematic viscosity, and the MDF-FDLBM will be more stable than that of IFDLBM with large Courant-Friedrichs-Lewy condition number. Finally, some simulations are conducted to validate the MDF-FDLBM. The results agree well with the analytical solutions and previous results. Through the numerical testing, we find that the MDF-FDLBM has a second-order convergence rate in space and time. The MDF-FDLBMcombined with non-uniform grid also works well. Meanwhile, compared with IFDLBM, it can be found that MDF-FDLBM offers higher accuracy and computational efficiency, reducing computation time by more than 36%.

physics.flu-dyn↗
arXiv2023

Numerical Simulation of Power-Law Fluid Flow in a Trapezoidal Cavity using the Incompressible Finite-Difference Lattice Boltzmann Method

In this paper, a numerical investigation of power-law fluid flow in the trapezoidal cavity has been conducted by incompressible finite-difference lattice Boltzmann method (IFDLBM). By designing the equilibrium distribution function, the Navier-Stokes equations (NSEs) can be recovered exactly. Through the coordinate transformation method, the body-fitted grid in physical region is transformed into a uniform grid in computational region. The effect of Reynolds (Re) number, the power-law index $n$ and the vertical angle θ on the trapezoidal cavity are investigated. According to the numerical results, we come to some conclusions. For low Re number Re=100, it can be found that the behavior of power-law fluid flow becomes more complicated with the increase of n. And as vertical angle θ decreases, the flow becomes smooth and the number of vortices decreases. For high Re numbers, the flow development becomes more complex, the number and strength of vortices increase. If the Reynolds number increases further, the power-law fluid will changes from steady flow to periodic flow and then to turbulent flow. For the steady flow, the lager the θ, the more complicated the vortices. And the critical Re number from steady to periodic state decreases with the decrease of power-law index n.

physics.flu-dyn↗
arXiv2023

AlphaCrystal: Contact map based crystal structure prediction using deep learning

Crystal structure prediction is one of the major unsolved problems in materials science. Traditionally, this problem is formulated as a global optimization problem for which global search algorithms are combined with first principle free energy calculations to predict the ground-state crystal structure given only a material composition or a chemical system. These ab initio algorithms usually cannot exploit a large amount of implicit physicochemical rules or geometric constraints (deep knowledge) of atom configurations embodied in a large number of known crystal structures. Inspired by the deep learning enabled breakthrough in protein structure prediction, herein we propose AlphaCrystal, a crystal structure prediction algorithm that combines a deep residual neural network model that learns deep knowledge to guide predicting the atomic contact map of a target crystal material followed by reconstructing its 3D crystal structure using genetic algorithms. Based on the experiments of a selected set of benchmark crystal materials, we show that our AlphaCrystal algorithm can predict structures close to the ground truth structures. It can also speed up the crystal structure prediction process by predicting and exploiting the predicted contact map so that it has the potential to handle relatively large systems. We believe that our deep learning based ab initio crystal structure prediction method that learns from existing material structures can be used to scale up current crystal structure prediction practice. To our knowledge, AlphaCrystal is the first neural network based algorithm for crystal structure contact map prediction and the first method for directly reconstructing crystal structures from materials composition, which can be further optimized by DFT calculations.

cond-mat.mtrl-sci↗
arXiv2023

Moments of proton GPDs from the OPE of nonlocal quark bilinears up to NNLO

For the first time, we present a lattice QCD determination of Mellin moments of unpolarized generalized parton distributions (GPDs) of the proton from an analysis of the quasi-GPD matrix elements within the short-distance factorization framework. We perform our calculation on an $N_f$=2+1+1 twisted mass fermions ensemble with a clover improvement at lattice spacing $a=0.093$ fm and a pion mass of $m_π=260$ MeV. Focusing on the zero-skewness case, the iso-vector and iso-scalar quasi-GPDs are calculated from the $γ_0$ definition, as well as a recently proposed Lorentz-invariant definition. We utilize data on both symmetric and asymmetric kinematic frames, which allows us to obtain the Mellin moments for several values of the momentum transfer, $-t$, in the range 0.17 to $2.77~\rm{GeV}^2$. We use the ratio scheme for GPDs, i.e. renormalization group invariant ratios with leading-twist factorization formula and perturbatively calculated matching coefficients up to the next-next-to-leading order (NNLO) to extract Mellin moments of GPDs, which are consistent with renormalization-group improved results. We compare our determination from quasi-GPDs with the results extracted using standard calculations of Mellin moments of local operators, specifically those related to the electromagnetic and gravitational form factors. We estimated the moments of GPDs up to the fifth ones for the first time. By extrapolating the Mellin moments to $-t=0$, we obtained the quark charges, momentum fraction, as well as the angular momentum contributions to the proton spin. The impact parameter space interpretation of the GPD moments is discussed, which provides insights into the spatial distribution of unpolarized quarks and their correlations in the transverse plane of an unpolarized or transversely polarized proton.

hep-lat↗
arXiv2023

Generalized Parton Distributions from Lattice QCD

In recent years, there has been a breakthrough in lattice calculations of $x$-dependent partonic distributions. This encompasses also distributions describing the 3D structure of the nucleon, such as generalized parton distributions (GPDs). We report a new method of accessing GPDs in asymmetric frames of reference, relying on a novel Lorentz-covariant parametrization of the accessed off-forward matrix elements in boosted nucleon states. The approach offers the possibility of computationally more efficient determination of the full parameter dependence of GPDs and as such, it can contribute to better understanding of nucleon's structure.

hep-lat↗
arXiv2023

Unpolarized proton PDF at NNLO from lattice QCD with physical quark masses

We present a lattice QCD calculation of the unpolarized isovector quark parton distribution function (PDF) of the proton utilizing a perturbative matching at next-to-next-to-leading-order (NNLO). The calculations are carried out using a single ensemble of gauge configurations generated with $N_f = 2 + 1$ highly-improved staggered quarks with physical masses and a lattice spacing of $a = 0.076$ fm. We use one iteration of hypercubic smearing on these gauge configurations, and the resulting smeared configurations are then used for all aspects of the subsequent calculation. For the valence quarks, we use the Wilson-clover action with physical quark masses. We consider several methods for extracting information on the PDF. We first extract the lowest four Mellin moments using the leading-twist operator product expansion approximation. Then, we determine the $x$ dependence of the PDF through a deep neural network within the pseudo-PDF approach and additionally through the framework of large-momentum effective theory utilizing a hybrid renormalization scheme. This is the first application of the NNLO matching coefficients for the nucleon directly at the physical point.

hep-lat↗
arXiv2023

TMD Handbook

This handbook provides a comprehensive review of transverse-momentum-dependent parton distribution functions and fragmentation functions, commonly referred to as transverse momentum distributions (TMDs). TMDs describe the distribution of partons inside the proton and other hadrons with respect to both their longitudinal and transverse momenta. They provide unique insight into the internal momentum and spin structure of hadrons, and are a key ingredient in the description of many collider physics cross sections. Understanding TMDs requires a combination of theoretical techniques from quantum field theory, nonperturbative calculations using lattice QCD, and phenomenological analysis of experimental data. The handbook covers a wide range of topics, from theoretical foundations to experimental analyses, as well as recent developments and future directions. It is intended to provide an essential reference for researchers and graduate students interested in understanding the structure of hadrons and the dynamics of partons in high energy collisions.

hep-ph↗
arXiv2023

FRA-LSTM: A Vessel Trajectory Prediction Method Based on Fusion of the Forward and Reverse Sub-Network

In order to improve the vessel's capacity and ensure maritime traffic safety, vessel intelligent trajectory prediction plays an essential role in the vessel's smart navigation and intelligent collision avoidance system. However, current researchers only focus on short-term or long-term vessel trajectory prediction, which leads to insufficient accuracy of trajectory prediction and lack of in-depth mining of comprehensive historical trajectory data. This paper proposes an Automatic Identification System (AIS) data-driven long short-term memory (LSTM) method based on the fusion of the forward sub-network and the reverse sub-network (termed as FRA-LSTM) to predict the vessel trajectory. The forward sub-network in our method combines LSTM and attention mechanism to mine features of forward historical trajectory data. Simultaneously, the reverse sub-network combines bi-directional LSTM (BiLSTM) and attention mechanism to mine features of backward historical trajectory data. Finally, the final predicted trajectory is generated by fusing output features of the forward and reverse sub-network. Based on plenty of experiments, we prove that the accuracy of our proposed method in predicting short-term and mid-term trajectories has increased by 96.8% and 86.5% on average compared with the BiLSTM and Seq2seq. Furthermore, the average accuracy of our method is 90.1% higher than that of compared the BiLSTM and Seq2seq in predicting long-term trajectories.

cs.CY↗
arXiv2023

Improving Transformer-based Networks With Locality For Automatic Speaker Verification

Recently, Transformer-based architectures have been explored for speaker embedding extraction. Although the Transformer employs the self-attention mechanism to efficiently model the global interaction between token embeddings, it is inadequate for capturing short-range local context, which is essential for the accurate extraction of speaker information. In this study, we enhance the Transformer with the enhanced locality modeling in two directions. First, we propose the Locality-Enhanced Conformer (LE-Confomer) by introducing depth-wise convolution and channel-wise attention into the Conformer blocks. Second, we present the Speaker Swin Transformer (SST) by adapting the Swin Transformer, originally proposed for vision tasks, into speaker embedding network. We evaluate the proposed approaches on the VoxCeleb datasets and a large-scale Microsoft internal multilingual (MS-internal) dataset. The proposed models achieve 0.75% EER on VoxCeleb 1 test set, outperforming the previously proposed Transformer-based models and CNN-based models, such as ResNet34 and ECAPA-TDNN. When trained on the MS-internal dataset, the proposed models achieve promising results with 14.6% relative reduction in EER over the Res2Net50 model.

eess.AS↗
arXiv2023

GPDs in asymmetric frames

It is often taken for granted that Generalized Parton Distributions (GPDs) are defined in the "symmetric" frame, where the transferred momentum is symmetrically distributed between the incoming/outgoing hadrons. However, such frames pose computational challenges for the lattice QCD practitioners. In these proceedings, we lay the foundation for lattice QCD calculations of GPDs in "asymmetric" frames, where the transferred momentum is not symmetrically distributed between the incoming/outgoing hadrons. The novelty of our work relies on the parameterization of the matrix elements in terms of Lorentz-invariant amplitudes, which not only helps in establishing relations between the said frames but also helps in isolating higher-twist contaminations. As an example, we focus on the unpolarized GPDs for spin-1/2 particles.

hep-lat↗
arXiv2023

Generalized Parton Distributions from Lattice QCD with Asymmetric Momentum Transfer: Unpolarized Quarks

Traditionally, lattice QCD computations of generalized parton distributions (GPDs) have been carried out in a symmetric frame, where the transferred momentum is symmetrically distributed between the incoming and outgoing hadrons. However, such frames are inconvenient since they require a separate calculation for each value of the momentum transfer, increasing significantly the computational cost. In this work, by focusing on the quasi-distribution approach, we lay the foundation for faster and more effective lattice QCD calculations of GPDs exploiting asymmetric frames, with freedom in the transferred momentum distribution. An important ingredient of our approach is the Lorentz covariant parameterization of the matrix elements in terms of Lorentz-invariant amplitudes, which allows one to relate matrix elements in different frames. We also use this amplitude approach to propose a new definition of quasi-GPDs that is frame-independent and, more importantly, may lead to smaller power corrections in the matching relations to the light-cone GPDs. We demonstrate the efficacy of the formalism through numerical calculations using one ensemble of $N_f$=2+1+1 twisted mass fermions with a clover improvement. The value of the light-quark masses lead to a pion mass of about 260 MeV. Concentrating on the proton, and limiting ourselves to a vanishing longitudinal momentum transfer to the target, we extract the invariant amplitudes from matrix element calculations in both the symmetric and asymmetric frame, and obtain results for the twist-2 light-cone GPDs for unpolarized quarks, that is, $H$ and $E$.

hep-lat↗
arXiv2022

Continuum-extrapolated NNLO Valence PDF of Pion at the Physical Point

We present lattice QCD calculations of valence parton distribution function (PDF) of pion employing next-to-next-leading-order (NNLO) perturbative QCD matching. Our calculations are based on three gauge ensembles of 2+1 flavor highly improved staggered quarks and Wilson--Clover valance quarks, corresponding to pion mass $m_π=140$~MeV at a lattice spacing $a=0.076$~fm and $m_π=300$~MeV at $a=0.04, 0.06$~fm. This enables us to present, for the first time, continuum-extrapolated lattice QCD results for NNLO valence PDF of the pion at the physical point. Applying leading-twist expansion for renormalization group invariant (RGI) ratios of bi-local pion matrix elements with NNLO Wilson coefficients we extract $2^{\mathrm{nd}}$, $4^{\mathrm{th}}$ and $6^{\mathrm{th}}$ Mellin moments of the PDF. We reconstruct the Bjorken-$x$ dependence of the NNLO PDF from real-space RGI ratios using a deep neural network (DNN) as well as from momentum-space matrix elements renormalized using a hybrid-scheme. All our results are in broad agreement with the results of global fits to the experimental data carried out by the xFitter and JAM collaborations.

hep-lat↗
arXiv2022

Accessing proton GPDs in asymmetric frames: Numerical implementation

In this work, we present a numerical investigation of a novel Lorentz-covariant parametrization to extract $x$-dependent GPDs using off-forward matrix elements of momentum-boosted hadrons coupled to non-local operators. The novelty of the method is the implementation of an asymmetric frame for the momentum transfer between the initial and final hadron state and the parametrization of the matrix elements into Lorentz-invariant amplitudes. The amplitudes can then be related to the standard light-cone GPDs. GPDs are defined in the symmetric frame, which requires a separate calculation for each value of the momentum transfer, increasing the computational cost significantly. The proposed method is powerful, as one can extract the GPDs at multiple values of the momentum transfer at the computational cost of a single value. For this proof-of-concept calculation, we use one ensemble of $N_f=2+1+1$ twisted mass fermions and a clover improvement with a pion mass of 260 MeV to calculate the proton unpolarized GPDs.

hep-lat↗
arXiv2022

Physics Guided Deep Learning for Generative Design of Crystal Materials with Symmetry Constraints

Discovering new materials is a challenging task in materials science crucial to the progress of human society. Conventional approaches based on experiments and simulations are labor-intensive or costly with success heavily depending on experts' heuristic knowledge. Here, we propose a deep learning based Physics Guided Crystal Generative Model (PGCGM) for efficient crystal material design with high structural diversity and symmetry. Our model increases the generation validity by more than 700\% compared to FTCP, one of the latest structure generators and by more than 45\% compared to our previous CubicGAN model. Density Functional Theory (DFT) calculations are used to validate the generated structures with 1,869 materials out of 2,000 are successfully optimized and deposited into the Carolina Materials Database \url{www.carolinamatdb.org}, of which 39.6\% have negative formation energy and 5.3\% have energy-above-hull less than 0.25 eV/atom, indicating their thermodynamic stability and potential synthesizability.

cond-mat.mtrl-sci↗
arXiv2022

Learnable Blur Kernel for Single-Image Defocus Deblurring in the Wild

Recent research showed that the dual-pixel sensor has made great progress in defocus map estimation and image defocus deblurring. However, extracting real-time dual-pixel views is troublesome and complex in algorithm deployment. Moreover, the deblurred image generated by the defocus deblurring network lacks high-frequency details, which is unsatisfactory in human perception. To overcome this issue, we propose a novel defocus deblurring method that uses the guidance of the defocus map to implement image deblurring. The proposed method consists of a learnable blur kernel to estimate the defocus map, which is an unsupervised method, and a single-image defocus deblurring generative adversarial network (DefocusGAN) for the first time. The proposed network can learn the deblurring of different regions and recover realistic details. We propose a defocus adversarial loss to guide this training process. Competitive experimental results confirm that with a learnable blur kernel, the generated defocus map can achieve results comparable to supervised methods. In the single-image defocus deblurring task, the proposed method achieves state-of-the-art results, especially significant improvements in perceptual quality, where PSNR reaches 25.56 dB and LPIPS reaches 0.111.

cs.CV↗
arXiv2022

Pion distribution amplitude at the physical point using the leading-twist expansion of the quasi-distribution-amplitude matrix element

We present a lattice QCD determination of the distribution amplitude (DA) of the pion and the first few Mellin moments from an analysis of the quasi-DA matrix element within the leading-twist framework. We perform our study on a HISQ ensemble with $a=0.076$ fm lattice spacing with the Wilson-Clover valence quark mass tuned to the physical point. We analyze the ratios of pion quasi-DA matrix elements at short distances using the leading-twist Mellin operator product expansion (OPE) at the next-to-leading order and the conformal OPE at the leading-logarithmic order. We find a robust result for the first non-vanishing Mellin moment $\langle x^2 \rangle = 0.287(6)(6)$ at a factorization scale $μ=2$ GeV. We also present different Ansätze-based reconstructions of the $x$-dependent DA, from which we determine the perturbative leading-twist expectations for the pion electromagnetic and gravitational form-factors at large momentum transfers.

hep-lat↗