arXiv subjects

Fei Jiang

Publications and source records attributed to Fei Jiang.

At least 73 records · Page 4Linked to original sources

arXiv2022

PANet: Perspective-Aware Network with Dynamic Receptive Fields and Self-Distilling Supervision for Crowd Counting

Crowd counting aims to learn the crowd density distributions and estimate the number of objects (e.g. persons) in images. The perspective effect, which significantly influences the distribution of data points, plays an important role in crowd counting. In this paper, we propose a novel perspective-aware approach called PANet to address the perspective problem. Based on the observation that the size of the objects varies greatly in one image due to the perspective effect, we propose the dynamic receptive fields (DRF) framework. The framework is able to adjust the receptive field by the dilated convolution parameters according to the input image, which helps the model to extract more discriminative features for each local region. Different from most previous works which use Gaussian kernels to generate the density map as the supervised information, we propose the self-distilling supervision (SDS) training method. The ground-truth density maps are refined from the first training stage and the perspective information is distilled to the model in the second stage. The experimental results on ShanghaiTech Part_A and Part_B, UCF_QNRF, and UCF_CC_50 datasets demonstrate that our proposed PANet outperforms the state-of-the-art methods by a large margin.

cs.CV↗

arXiv2022

CGMN: A Contrastive Graph Matching Network for Self-Supervised Graph Similarity Learning

Graph similarity learning refers to calculating the similarity score between two graphs, which is required in many realistic applications, such as visual tracking, graph classification, and collaborative filtering. As most of the existing graph neural networks yield effective graph representations of a single graph, little effort has been made for jointly learning two graph representations and calculating their similarity score. In addition, existing unsupervised graph similarity learning methods are mainly clustering-based, which ignores the valuable information embodied in graph pairs. To this end, we propose a contrastive graph matching network (CGMN) for self-supervised graph similarity learning in order to calculate the similarity between any two input graph objects. Specifically, we generate two augmented views for each graph in a pair respectively. Then, we employ two strategies, namely cross-view interaction and cross-graph interaction, for effective node representation learning. The former is resorted to strengthen the consistency of node representations in two views. The latter is utilized to identify node differences between different graphs. Finally, we transform node representations into graph-level representations via pooling operations for graph similarity computation. We have evaluated CGMN on eight real-world datasets, and the experiment results show that the proposed new approach is superior to the state-of-the-art methods in graph similarity learning downstream tasks.

cs.LG↗

arXiv2022

Student Dangerous Behavior Detection in School

Video surveillance systems have been installed to ensure the student safety in schools. However, discovering dangerous behaviors, such as fighting and falling down, usually depends on untimely human observations. In this paper, we focus on detecting dangerous behaviors of students automatically, which faces numerous challenges, such as insufficient datasets, confusing postures, keyframes detection and prompt response. To address these challenges, we first build a danger behavior dataset with locations and labels from surveillance videos, and transform action recognition of long videos to an object detection task that avoids keyframes detection. Then, we propose a novel end-to-end dangerous behavior detection method, named DangerDet, that combines multi-scale body features and keypoints-based pose features. We could improve the accuracy of behavior classification due to the highly correlation between pose and behavior. On our dataset, DangerDet achieves 71.0\% mAP with about 11 FPS. It keeps a better balance between the accuracy and time cost.

cs.CV↗

arXiv2022

On Inhibition of Rayleigh--Taylor Instability by a Horizontal Magnetic Field in Non-resistive MHD Fluids: the Viscous Case

It is still open whether the phenomenon of inhibition of Rayleigh--Taylor (RT) instability by a horizontal magnetic field can be mathematically verified for a non-resistive \emph{viscous} magnetohydrodynamic (MHD) fluid in a two-dimensional (2D) horizontal slab domain, since it was roughly proved in the linearized case by Wang in \cite{WYC}. In this paper, we prove such inhibition phenomenon by the (nonlinear) inhomogeneous, incompressible, \emph{viscous case} with \emph{Navier (slip) boundary condition}. More precisely, we show that there is a critical number of field strength $m_{\mm{C}}$, such that if the strength $|m|$ of a horizontal magnetic field is bigger than $m_{\mm{C}}$, then the small perturbation solution around the magnetic RT equilibrium state is {algebraically} stable in time. In addition, we also provide a nonlinear instability result for the case $|m|\in[0, m_{\mm{C}})$. The instability result presents that a horizontal magnetic field can not inhibit the RT instability, if it's strength is too small.

math.AP↗

arXiv2021

On Magnetic Inhibition Theory in Non-resistive Magnetohydrodynamic Fluids: Existence of Solutions in Some Classes of Large Data

This paper is concerned with existence of solutions to the incompressible non-resistive viscous magnetohydrodynamic (MHD) equations with large initial perturbations in there-dimensional (3D) periodic domains (in Lagrangian coordinates). Motivated by the Diophantine condition imposed by the approximate theory of non-resistive MHD equations in \cite{BCSCSPLL}, Chen--Zhang--Zhou in \cite{chen2021} and the magnetic inhibition mechanism of Lagrangian coordinates version in our previous paper \cite{JFJSOMITIN}, we prove the existence of unique classical solutions under some class of large initial perturbations, where the intensity of impressive magnetic fields depends increasingly on the $ H^{17}\times H^{21}$-norm of the initial perturbation of both the velocity and magnetic field. Our result not only mathematically verifies that magnetic fields prevent the singularity formation of solutions with large initial velocity in the viscous case, but also provide a starting point for the existence theory of large perturbation solutions of the 3D non-resistive viscous MHD equations. In addition, we further rigorously prove that, for large time or strong magnetic field, the MHD equations reduce to the corresponding linearized equations by providing the error estimates, which enjoy the types of algebraic decay with respect to the both of time and field intensity, between the solutions of both the nonlinear and linear equations.

math.AP↗

arXiv2021

An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems

Spoofing countermeasure (CM) systems are critical in speaker verification; they aim to discern spoofing attacks from bona fide speech trials. In practice, however, acoustic condition variability in speech utterances may significantly degrade the performance of CM systems. In this paper, we conduct a cross-dataset study on several state-of-the-art CM systems and observe significant performance degradation compared with their single-dataset performance. Observing differences of average magnitude spectra of bona fide utterances across the datasets, we hypothesize that channel mismatch among these datasets is one important reason. We then verify it by demonstrating a similar degradation of CM systems trained on original but evaluated on channel-shifted data. Finally, we propose several channel robust strategies (data augmentation, multi-task learning, adversarial learning) for CM systems, and observe a significant performance improvement on cross-dataset experiments.

eess.AS↗

arXiv2021

Y-Vector: Multiscale Waveform Encoder for Speaker Embedding

State-of-the-art text-independent speaker verification systems typically use cepstral features or filter bank energies as speech features. Recent studies attempted to extract speaker embeddings directly from raw waveforms and have shown competitive results. In this paper, we propose a novel multi-scale waveform encoder that uses three convolution branches with different time scales to compute speech features from the waveform. These features are then processed by squeeze-and-excitation blocks, a multi-level feature aggregator, and a time delayed neural network (TDNN) to compute speaker embedding. We show that the proposed embeddings outperform existing raw-waveform-based speaker embeddings on speaker verification by a large margin. A further analysis of the learned filters shows that the multi-scale encoder attends to different frequency bands at its different scales while resulting in a more flat overall frequency response than any of the single-scale counterparts.

eess.AS↗

arXiv2021

Deoscillated Graph Collaborative Filtering

Collaborative Filtering (CF) signals are crucial for a Recommender System~(RS) model to learn user and item embeddings. High-order information can alleviate the cold-start issue of CF-based methods, which is modelled through propagating the information over the user-item bipartite graph. Recent Graph Neural Networks~(GNNs) propose to stack multiple aggregation layers to propagate high-order signals. However, the oscillation problem, varying locality of bipartite graph, and the fix propagation pattern spoil the ability of multi-layer structure to propagate information. The oscillation problem results from the bipartite structure, as the information from users only propagates to items. Besides oscillation problem, varying locality suggests the density of nodes should be considered in the propagation process. Moreover, the layer-fixed propagation pattern introduces redundant information between layers. In order to tackle these problems, we propose a new RS model, named as \textbf{D}eoscillated \textbf{G}raph \textbf{C}ollaborative \textbf{F}iltering~(DGCF). We introduce cross-hop propagation layers in it to break the bipartite propagating structure, thus resolving the oscillation problem. Additionally, we design innovative locality-adaptive layers which adaptively propagate information. Stacking multiple cross-hop propagation layers and locality layers constitutes the DGCF model, which models high-order CF signals adaptively to the locality of nodes and layers. Extensive experiments on real-world datasets show the effectiveness of DGCF. Detailed analyses indicate that DGCF solves oscillation problem, adaptively learns local factor, and has layer-wise propagation pattern. Our code is available online at https://github.com/JimLiu96/DeosciRec.

cs.IR↗

arXiv2021

Global Solutions of Three-dimensional Inviscid MHD Fluids with Velocity Damping in Horizontally Periodic Domains

The \emph{two-dimensional} (2D) existence result of global(-in-time) solutions for the motion equations of incompressible, inviscid, non-resistive magnetohydrodynamic (MHD) fluids with velocity damping had been established in [Wu--Wu--Xu, SIAM J. Math. Anal. 47 (2013), 2630--2656]. This paper further studies the existence of global solutions for the \emph{three-dimensional} (a dimension of real world) initial-boundary value problem in a horizontally periodic domain with finite height. Motivated by the multi-layers energy method introduced in [Guo--Tice, Arch. Ration. Mech. Anal. 207 (2013), 459--531], we develop a new type of two-layer energy structure to overcome the difficulty arising from three-dimensional nonlinear terms in the MHD equations, and thus prove the initial-boundary value problem admits a unique global solution. Moreover the solution has the exponential decay-in-time around some rest state. Our two-layer energy structure enjoys two features: (1) the lower-order energy (functional) can not be controlled by the higher-order energy. (2) under the \emph{a priori} smallness assumption of lower-order energy, we first close the higher-order energy estimates, and then further close the lower-energy estimates in turn.

math.AP↗

arXiv2021

On Inhibition of Rayleigh--Taylor Instability by Horizontal Magnetic Field in an Inviscid MHD Fluid with Velocity Damping

It is still an open problem whether the inhibition phenomenon of Rayleigh--Taylor (RT) instability by horizontal magnetic field can be mathematically proved in a non-resistive magnetohydrodynamic (MHD) fluid in a two-dimensional (2D) horizontal slab domain, since it had been roughly verified by a 2D linearized motion equations in 2012 \cite{WYC}. In this paper, we find that this inhibition phenomenon can be rigorously verified in the inhomogeneous, incompressible, inviscid case with velocity damping. More precisely, there exists a critical number $m_{\rm{C}}$ such that if the strength $|m|$ of horizontal magnetic field is bigger than $m_{\rm{C}}$, then the small perturbation solution around the magnetic RT equilibrium state is exponentially stable in time. Our result is also the first mathematical one based on the nonlinear motion equations for the proof of inhibition of flow instabilities by a horizontal magnetic field in a horizontal slab domain. In addition, we also provide a nonlinear instability result for the case $|m|\in [0,m_{\rm{C}})$. Our instability result presents that horizontal magnetic field can not inhibit the RT instability, if it's strength is to small.

math.AP↗

arXiv2021

Prediction of three-phase relative permeabilities of Berea sandstone using lattice Boltzmann method

Three-phase flows through a pore network of Berea sandstone are studied numerically under critical interfacial tension condition. Results show that the relative permeability of each fluid increases as its own saturation increases. The specific interfacial length between wetting and non-wetting fluids monotonously decreases with increasing the saturation of intermediate-wetting fluid, while the other two specific interfacial lengths exhibit a non-monotonous variation. As the wetting (non-wetting) fluid becomes less wetting (non-wetting), the relative permeability of wetting fluid monotonously increases, while the other two relative permeabilities show a non-monotonous trend. Due to the presence of spreading layer, the specific interfacial length between wetting and non-wetting fluids always stabilizes at a low level. As the viscosity ratio of wetting (non-wetting) to intermediate-wetting fluids increases, the relative permeability of wetting (non-wetting) fluid increases. With the viscosity ratio deviating from unity, the phase interfaces become increasingly unstable, leading to an increased specific interfacial length.

physics.flu-dyn↗

arXiv2021

Asymptotic Behaviors of Global Solutions to the Two-Dimensional Non-resistive MHD Equations with Large Initial Perturbations

This paper is concerned with the asymptotic behaviors of global strong solutions to the incompressible non-resistive viscous magnetohydrodynamic (MHD) equations with large initial perturbations in two-dimensional periodic domains in Lagrangian coordinates. First, motivated by the odevity conditions imposed in [Arch. Ration. Mech. Anal. 227 (2018), 637--662], we prove the existence and uniqueness of strong solutions under some class of large initial perturbations, where the strength of impressive magnetic fields depends increasingly on the $H^2$-norm of the initial perturbation values of both velocity and magnetic field. Then, we establish time-decay rates of strong solutions. Moreover, we find that $H^2$-norm of the velocity decays faster than the perturbed magnetic field. Finally, by developing some new analysis techniques, we show that the strong solution convergence in a rate of the field strength to the solution of the corresponding linearized problem as the strength of the impressive magnetic field goes to infinity. In addition, an extension of similar results to the corresponding inviscid case with damping is presented.

math.AP↗

arXiv2021

Lattice Boltzmann simulation of three-phase flows with moving contact lines on curved surfaces

A numerical method for simulating three-phase flows with moving contact lines on arbitrarily complex surfaces is developed in the framework of lattice Boltzmann method. In this method, the immiscible three-phase flow is modeled through a multiple-relaxation-time color-gradient model, which not only allows for a full range of interfacial tensions but also produces stable outcomes for a wide range of viscosity ratios. A characteristic line model is introduced to implement the wetting boundary condition, which is not only easy to implement but also able to handle arbitrarily complex boundaries with prescribed contact angles. The developed method is first validated by the simulation of a Janus droplet resting on a flat surface, a perfect Janus droplet deposited on a cylinder, and the capillary intrusion of ternary fluids for various viscosity ratios. It is then used to study a compound droplet subject to a uniform incoming flow passing through a multi-pillar structure, where three different values of surface wettability are considered. The simulated results show that the surface wettability has significant impact on the droplet dynamic behavior and final fluid distribution.

physics.flu-dyn↗

arXiv2021

One-class Learning Towards Synthetic Voice Spoofing Detection

Human voices can be used to authenticate the identity of the speaker, but the automatic speaker verification (ASV) systems are vulnerable to voice spoofing attacks, such as impersonation, replay, text-to-speech, and voice conversion. Recently, researchers developed anti-spoofing techniques to improve the reliability of ASV systems against spoofing attacks. However, most methods encounter difficulties in detecting unknown attacks in practical use, which often have different statistical distributions from known attacks. Especially, the fast development of synthetic voice spoofing algorithms is generating increasingly powerful attacks, putting the ASV systems at risk of unseen attacks. In this work, we propose an anti-spoofing system to detect unknown synthetic voice spoofing attacks (i.e., text-to-speech or voice conversion) using one-class learning. The key idea is to compact the bona fide speech representation and inject an angular margin to separate the spoofing attacks in the embedding space. Without resorting to any data augmentation methods, our proposed system achieves an equal error rate (EER) of 2.19% on the evaluation set of ASVspoof 2019 Challenge logical access scenario, outperforming all existing single systems (i.e., those without model ensemble).

eess.AS↗

arXiv2020

Strong Solutions of the Equations for Viscoelastic Fluids in Some Classes of Large Data

We study the existence and uniqueness of global strong solutions to the equations of an incompressible viscoelastic fluid in a spatially periodic domain, and show that a unique strong solution exists globally in time if the initial deformation and velocity are small for the given physical parameters. In particular, the initial velocity can be large for the large elasticity coefficient. The result of this paper mathematically verifies that the elasticity can prevent the formation of singularities of strong solutions with large initial velocity, thus playing a similar role to viscosity in preventing the formation of singularities in viscous flows. Moreover, for given initial velocity perturbation and zero initial deformation around the rest state, we find, as the elasticity coefficient or time go to infinity, that (1) any straight line segment $l^0$ consisted of fluid particles in the rest state, after being bent by a velocity perturbation, will turn into a straight line segment that is parallel to $l^0$ and has the same length as $l^0$. (2) the motion of the viscoelastic fluid can be approximated by a linear pressureless motion in Lagrangian coordinates, even when the initial velocity is large. Moreover, the above mentioned phenomena can also be found in the corresponding compressible fluid case.

math.AP↗

arXiv2019

Community-preserving Graph Convolutions for Structural and Functional Joint Embedding of Brain Networks

Brain networks have received considerable attention given the critical significance for understanding human brain organization, for investigating neurological disorders and for clinical diagnostic applications. Structural brain network (e.g. DTI) and functional brain network (e.g. fMRI) are the primary networks of interest. Most existing works in brain network analysis focus on either structural or functional connectivity, which cannot leverage the complementary information from each other. Although multi-view learning methods have been proposed to learn from both networks (or views), these methods aim to reach a consensus among multiple views, and thus distinct intrinsic properties of each view may be ignored. How to jointly learn representations from structural and functional brain networks while preserving their inherent properties is a critical problem. In this paper, we propose a framework of Siamese community-preserving graph convolutional network (SCP-GCN) to learn the structural and functional joint embedding of brain networks. Specifically, we use graph convolutions to learn the structural and functional joint embedding, where the graph structure is defined with structural connectivity and node features are from the functional connectivity. Moreover, we propose to preserve the community structure of brain networks in the graph convolutions by considering the intra-community and inter-community properties in the learning process. Furthermore, we use Siamese architecture which models the pair-wise similarity learning to guide the learning process. To evaluate the proposed approach, we conduct extensive experiments on two real brain network datasets. The experimental results demonstrate the superior performance of the proposed approach in structural and functional joint embedding for neurological disorder analysis, indicating its promising value for clinical applications.

cs.LG↗

arXiv2019

WaveletAE: A Wavelet-enhanced Autoencoder for Wind Turbine Blade Icing Detection

Wind power, as an alternative to burning fossil fuels, is abundant and inexhaustible. To fully utilize wind power, wind farms are usually located in areas of high altitude and facing serious ice conditions, which can lead to serious consequences. Quick detection of blade ice accretion is crucial for the maintenance of wind farms. Unlike traditional methods of installing expensive physical detectors on wind blades, data-driven approaches are increasingly popular for inspecting the wind turbine failures. In this work, we propose a wavelet enhanced autoencoder model (WaveletAE) to identify wind turbine dysfunction by analyzing the multivariate time series monitored by the SCADA system. WaveletAE is enhanced with wavelet detail coefficients to enforce the autoencoder to capture information from multiple scales, and the CNN-LSTM architecture is applied to learn channel-wise and temporal-wise relations. The empirical study shows that the proposed model outperforms other state-of-the-art time series anomaly detection methods for real-world blade icing detection.

cs.LG↗

arXiv2019

A spline-assisted semiparametric approach to non-parametric measurement error models

It is well known that the minimax rates of convergence of nonparametric density and regression function estimation of a random variable measured with error is much slower than the rate in the error free case. Surprisingly, we show that if one is willing to impose a relatively mild assumption in requiring that the error-prone variable has a compact support, then the results can be greatly improved. We describe new and constructive methods to take full advantage of the compact support assumption via spline-assisted semiparametric methods. We further prove that the new estimator achieves the usual nonparametric rate in estimating both the density and regression functions as if there were no measurement error. The proof involves linear and bilinear operator theories, semiparametric theory, asymptotic analysis regarding Bsplines, as well as integral equation treatments. The performance of the new methods is demonstrated through several simulations and a data example.

math.ST↗