arXiv subjects

Qiang Wu

Publications and source records attributed to Qiang Wu.

At least 127 records · Page 7Linked to original sources

arXiv2021

Fluctuation results for Multi-species Sherrington-Kirkpatrick model in the replica symmetric regime

We study the Replica Symmetric region of general multi-species Sherrington-Kirkpatrick (MSK) Model and answer some of the questions raised in Ann.~Probab.~43~(2015), no.~6, 3494--3513, where the author proved the Parisi formula under \emph{positive-definite} assumption on the disorder covariance matrix $Δ^2$. First, we prove exponential overlap concentration at high temperature for both \indf~and \emph{positive-definite} $Δ^2$ MSK model. We also prove a central limit theorem for the free energy using overlap concentration. Furthermore, in the zero external field case, we use a quadratic coupling argument to prove overlap concentration up to $β_c$, which is expected to be the critical inverse temperature. The argument holds for both \emph{positive-definite} and emph{indefinite} $Δ^2$, and $β_c$ has the same expression in two different cases. Second, we develop a species-wise cavity approach to study the overlap fluctuation, and the asymptotic variance-covariance matrix of overlap is obtained as the solution to a matrix-valued linear system. The asymptotic variance also suggests the de Almeida--Thouless (AT) line condition from the Replica Symmetry (RS) side. Our species-wise cavity approach does not require the positive-definiteness of $Δ^2$. However, it seems that the AT line conditions in \emph{positive-definite} and \emph{indefinite} cases are different. Finally, in the case of \emph{positive-definite} $Δ^2$, we prove that above the AT line, the MSK model is in Replica Symmetry Breaking phase under some natural assumption. This generalizes the results of J.~Stat.~Phys.~174 (2019), no.~2, 333--350, from 2-species to general species.

math.PR↗

arXiv2021

Approximation algorithms for the random-field Ising model

Approximating the partition function of the ferromagnetic Ising model with general external fields is known to be #BIS-hard in the worst case, even for bounded-degree graphs, and it is widely believed that no polynomial-time approximation scheme exists. This motivates an average-case question: are there classes of instances for which polynomial-time approximation schemes exist? We investigate this question for the random field Ising model on graphs with maximum degree $Δ$. We establish the existence of fully polynomial-time approximation schemes and samplers with high probability over the random fields if the external fields are IID Gaussians with variance larger than a constant depending only on the inverse temperature and $Δ$. The main challenge comes from the positive density of vertices at which the external field is small. These regions, which may have connected components of size $Θ(\log n)$, are a barrier to algorithms based on establishing a zero-free region, and cause worst-case analyses of Glauber dynamics to fail. The analysis of our algorithm is based on percolation on a self-avoiding walk tree.

cs.DS↗

arXiv2021

TVDIM: Enhancing Image Self-Supervised Pretraining via Noisy Text Data

Among ubiquitous multimodal data in the real world, text is the modality generated by human, while image reflects the physical world honestly. In a visual understanding application, machines are expected to understand images like human. Inspired by this, we propose a novel self-supervised learning method, named Text-enhanced Visual Deep InfoMax (TVDIM), to learn better visual representations by fully utilizing the naturally-existing multimodal data. Our core idea of self-supervised learning is to maximize the mutual information between features extracted from multiple views of a shared context to a rational degree. Different from previous methods which only consider multiple views from a single modality, our work produces multiple views from different modalities, and jointly optimizes the mutual information for features pairs of intra-modality and inter-modality. Considering the information gap between inter-modality features pairs from data noise, we adopt a \emph{ranking-based} contrastive learning to optimize the mutual information. During evaluation, we directly use the pre-trained visual representations to complete various image classification tasks. Experimental results show that, TVDIM significantly outperforms previous visual self-supervised methods when processing the same set of images.

cs.CL↗

arXiv2021

InfoBehavior: Self-supervised Representation Learning for Ultra-long Behavior Sequence via Hierarchical Grouping

E-commerce companies have to face abnormal sellers who sell potentially-risky products. Typically, the risk can be identified by jointly considering product content (e.g., title and image) and seller behavior. This work focuses on behavior feature extraction as behavior sequences can provide valuable clues for the risk discovery by reflecting the sellers' operation habits. Traditional feature extraction techniques heavily depend on domain experts and adapt poorly to new tasks. In this paper, we propose a self-supervised method InfoBehavior to automatically extract meaningful representations from ultra-long raw behavior sequences instead of the costly feature selection procedure. InfoBehavior utilizes Bidirectional Transformer as feature encoder due to its excellent capability in modeling long-term dependency. However, it is intractable for commodity GPUs because the time and memory required by Transformer grow quadratically with the increase of sequence length. Thus, we propose a hierarchical grouping strategy to aggregate ultra-long raw behavior sequences to length-processable high-level embedding sequences. Moreover, we introduce two types of pretext tasks. Sequence-related pretext task defines a contrastive-based training objective to correctly select the masked-out coarse-grained/fine-grained behavior sequences against other "distractor" behavior sequences; Domain-related pretext task designs a classification training objective to correctly predict the domain-specific statistical results of anomalous behavior. We show that behavior representations from the pre-trained InfoBehavior can be directly used or integrated with features from other side information to support a wide range of downstream tasks. Experimental results demonstrate that InfoBehavior significantly improves the performance of Product Risk Management and Intellectual Property Protection.

cs.CL↗

arXiv2021

Dual Attention on Pyramid Feature Maps for Image Captioning

Generating natural sentences from images is a fundamental learning task for visual-semantic understanding in multimedia. In this paper, we propose to apply dual attention on pyramid image feature maps to fully explore the visual-semantic correlations and improve the quality of generated sentences. Specifically, with the full consideration of the contextual information provided by the hidden state of the RNN controller, the pyramid attention can better localize the visually indicative and semantically consistent regions in images. On the other hand, the contextual information can help re-calibrate the importance of feature components by learning the channel-wise dependencies, to improve the discriminative power of visual features for better content description. We conducted comprehensive experiments on three well-known datasets: Flickr8K, Flickr30K and MS COCO, which achieved impressive results in generating descriptive and smooth natural sentences from images. Using either convolution visual features or more informative bottom-up attention features, our composite captioning model achieves very promising performance in a single-model mode. The proposed pyramid attention and dual attention methods are highly modular, which can be inserted into various image captioning modules to further improve the performance.

cs.CV↗

arXiv2021

Modelling the Sgr A* Black Hole Immersed in a Dark Matter Spike

In this paper, we investigate the effects of a dark matter (DM) spike on the neighborhood of Sgr A*, the black hole (BH) in the center of the Milky Way galaxy. Our main goal is to investigate whether current and future astronomical observations of Sgr A* could detect the presence of such a DM spike. At first, we construct the spacetime metric around a static and spherically symmetric BH with a DM spike, and later this solution is generalized for a rotating BH using the Newman-Janis-Azreg Aïnou algorithm. For the static BH metric, we use the data of the S2 star orbiting the Sgr A* to determine and analyze the constraints on the two free parameters characterizing the density and the innermost boundary of the DM halo surrounding the BH. Furthermore, by making use of the available observational data for the DM spike density $ρ_\text{sp}$ and the DM spike radius $R_\text{sp}$ in the Milky Way galaxy, we consider a geometrically-thick accretion disk model around the Sgr A* BH and demonstrate that the effect of DM distribution on the shadow radius and the image of the BH is considerably weak for realistic DM densities, becoming significant only when the DM density is of the order $ρ_\text{sp} \sim (10^{-19}-10^{-20})$ g/cm$^3$ near the BH. We further analyze the possibility of observing this effect with radio interferometry, simulating observations with an EHT--like array, and find that it is unlikely to be detectable in the near future.

astro-ph.HE↗

arXiv2021

Shadow and Quasinormal Modes of a Rotating Loop Quantum Black Hole

In this paper, we construct an effective rotating loop quantum black hole (LQBH) solution, starting from the spherical symmetric LQBH by applying the Newman-Janis algorithm modified by Azreg-Aïnou's non-complexification procedure, and study the effects of loop quantum gravity { (LQG) on its shadow}. Given the rotating {LQBH}, we discuss its horizon, ergosurface, and regularity {as} $r \to 0$. Depending on the values of the specific angular momentum $a$ and the polymeric function $P$ arising from {LQG}, we {find} that the rotating solution we obtained can represent a regular black hole, a regular extreme black hole, or a regular spacetime {without horizon (a non-black-hole solution)}. We also {study} the effects of {LQG} and rotation, and {show} that, in addition to the specific angular momentum, the polymeric function {also} causes deformations in the size and shape of the black hole shadow. Interestingly, for a given value of $a$ and inclination angle $θ_0$, the apparent size of the shadow monotonically decreases, and the shadow gets more distorted with increasing $P$. We also {consider the effects of $P$ on the deviations from the circularity of the shadow, and find} that the deviation from circularity increases with increasing $P$ for fixed values of $a$ and $θ_0$. Additionally, we explore the observational implications of $P$ in comparison with the latest Event Horizon Telescope (EHT) observation of the supermassive black hole, M$87$. The connection between the shadow radius and quasinormal modes in the eikonal limit as well as {the} deflection of massive particles are also considered.

gr-qc↗

arXiv2021

PTN: A Poisson Transfer Network for Semi-supervised Few-shot Learning

The predicament in semi-supervised few-shot learning (SSFSL) is to maximize the value of the extra unlabeled data to boost the few-shot learner. In this paper, we propose a Poisson Transfer Network (PTN) to mine the unlabeled information for SSFSL from two aspects. First, the Poisson Merriman Bence Osher (MBO) model builds a bridge for the communications between labeled and unlabeled examples. This model serves as a more stable and informative classifier than traditional graph-based SSFSL methods in the message-passing process of the labels. Second, the extra unlabeled samples are employed to transfer the knowledge from base classes to novel classes through contrastive learning. Specifically, we force the augmented positive pairs close while push the negative ones distant. Our contrastive transfer scheme implicitly learns the novel-class embeddings to alleviate the over-fitting problem on the few labeled data. Thus, we can mitigate the degeneration of embedding generality in novel classes. Extensive experiments indicate that PTN outperforms the state-of-the-art few-shot and SSFSL models on miniImageNet and tieredImageNet benchmark datasets.

cs.CV↗

arXiv2021

TOAN: Target-Oriented Alignment Network for Fine-Grained Image Categorization with Few Labeled Samples

The challenges of high intra-class variance yet low inter-class fluctuations in fine-grained visual categorization are more severe with few labeled samples, \textit{i.e.,} Fine-Grained categorization problems under the Few-Shot setting (FGFS). High-order features are usually developed to uncover subtle differences between sub-categories in FGFS, but they are less effective in handling the high intra-class variance. In this paper, we propose a Target-Oriented Alignment Network (TOAN) to investigate the fine-grained relation between the target query image and support classes. The feature of each support image is transformed to match the query ones in the embedding feature space, which reduces the disparity explicitly within each category. Moreover, different from existing FGFS approaches devise the high-order features over the global image with less explicit consideration of discriminative parts, we generate discriminative fine-grained features by integrating compositional concept representations to global second-order pooling. Extensive experiments are conducted on four fine-grained benchmarks to demonstrate the effectiveness of TOAN compared with the state-of-the-art models.

cs.CV↗

arXiv2021

ShadowSync: Performing Synchronization in the Background for Highly Scalable Distributed Training

Recommendation systems are often trained with a tremendous amount of data, and distributed training is the workhorse to shorten the training time. While the training throughput can be increased by simply adding more workers, it is also increasingly challenging to preserve the model quality. In this paper, we present \shadowsync, a distributed framework specifically tailored to modern scale recommendation system training. In contrast to previous works where synchronization happens as part of the training process, \shadowsync separates the synchronization from training and runs it in the background. Such isolation significantly reduces the synchronization overhead and increases the synchronization frequency, so that we are able to obtain both high throughput and excellent model quality when training at scale. The superiority of our procedure is confirmed by experiments on training deep neural networks for click-through-rate prediction tasks. Our framework is capable to express data parallelism and/or model parallelism, generic to host various types of synchronization algorithms, and readily applicable to large scale problems in other areas.

cs.LG↗

arXiv2021

A Framework of Learning Through Empirical Gain Maximization

We develop in this paper a framework of empirical gain maximization (EGM) to address the robust regression problem where heavy-tailed noise or outliers may present in the response variable. The idea of EGM is to approximate the density function of the noise distribution instead of approximating the truth function directly as usual. Unlike the classical maximum likelihood estimation that encourages equal importance of all observations and could be problematic in the presence of abnormal observations, EGM schemes can be interpreted from a minimum distance estimation viewpoint and allow the ignorance of those observations. Furthermore, it is shown that several well-known robust nonconvex regression paradigms, such as Tukey regression and truncated least square regression, can be reformulated into this new framework. We then develop a learning theory for EGM, by means of which a unified analysis can be conducted for these well-established but not fully-understood regression approaches. Resulting from the new framework, a novel interpretation of existing bounded nonconvex loss functions can be concluded. Within this new framework, the two seemingly irrelevant terminologies, the well-known Tukey's biweight loss for robust regression and the triweight kernel for nonparametric smoothing, are closely related. More precisely, it is shown that the Tukey's biweight loss can be derived from the triweight kernel. Similarly, other frequently employed bounded nonconvex loss functions in machine learning such as the truncated square loss, the Geman-McClure loss, and the exponential squared loss can also be reformulated from certain smoothing kernels in statistics. In addition, the new framework enables us to devise new bounded nonconvex loss functions for robust learning.

cs.LG↗

arXiv2020

Field-wise Learning for Multi-field Categorical Data

We propose a new method for learning with multi-field categorical data. Multi-field categorical data are usually collected over many heterogeneous groups. These groups can reflect in the categories under a field. The existing methods try to learn a universal model that fits all data, which is challenging and inevitably results in learning a complex model. In contrast, we propose a field-wise learning method leveraging the natural structure of data to learn simple yet efficient one-to-one field-focused models with appropriate constraints. In doing this, the models can be fitted to each category and thus can better capture the underlying differences in data. We present a model that utilizes linear models with variance and low-rank constraints, to help it generalize better and reduce the number of parameters. The model is also interpretable in a field-wise manner. As the dimensionality of multi-field categorical data can be very high, the models applied to such data are mostly over-parameterized. Our theoretical analysis can potentially explain the effect of over-parametrization on the generalization of our model. It also supports the variance constraints in the learning objective. The experiment results on two large-scale datasets show the superior performance of our model, the trend of the generalization error bound, and the interpretability of learning outcomes. Our code is available at https://github.com/lzb5600/Field-wise-Learning.

cs.LG↗

arXiv2020

Multi-layer Feature Aggregation for Deep Scene Parsing Models

Scene parsing from images is a fundamental yet challenging problem in visual content understanding. In this dense prediction task, the parsing model assigns every pixel to a categorical label, which requires the contextual information of adjacent image patches. So the challenge for this learning task is to simultaneously describe the geometric and semantic properties of objects or a scene. In this paper, we explore the effective use of multi-layer feature outputs of the deep parsing networks for spatial-semantic consistency by designing a novel feature aggregation module to generate the appropriate global representation prior, to improve the discriminative power of features. The proposed module can auto-select the intermediate visual features to correlate the spatial and semantic information. At the same time, the multiple skip connections form a strong supervision, making the deep parsing network easy to train. Extensive experiments on four public scene parsing datasets prove that the deep parsing network equipped with the proposed feature aggregation module can achieve very promising results.

cs.CV↗

arXiv2020

A Statistical Learning Assessment of Huber Regression

As one of the triumphs and milestones of robust statistics, Huber regression plays an important role in robust inference and estimation. It has also been finding a great variety of applications in machine learning. In a parametric setup, it has been extensively studied. However, in the statistical learning context where a function is typically learned in a nonparametric way, there is still a lack of theoretical understanding of how Huber regression estimators learn the conditional mean function and why it works in the absence of light-tailed noise assumptions. To address these fundamental questions, we conduct an assessment of Huber regression from a statistical learning viewpoint. First, we show that the usual risk consistency property of Huber regression estimators, which is usually pursued in machine learning, cannot guarantee their learnability in mean regression. Second, we argue that Huber regression should be implemented in an adaptive way to perform mean regression, implying that one needs to tune the scale parameter in accordance with the sample size and the moment condition of the noise. Third, with an adaptive choice of the scale parameter, we demonstrate that Huber regression estimators can be asymptotic mean regression calibrated under $(1+ε)$-moment conditions ($ε>0$). Last but not least, under the same moment conditions, we establish almost sure convergence rates for Huber regression estimators. Note that the $(1+ε)$-moment conditions accommodate the special case where the response variable possesses infinite variance and so the established convergence rates justify the robustness feature of Huber regression estimators. In the above senses, the present study provides a systematic statistical learning assessment of Huber regression estimators and justifies their merits in terms of robustness from a theoretical viewpoint.

math.ST↗

arXiv2020

Thin Accretion Disk around a four-dimensional Einstein-Gauss-Bonnet Black Hole

Recently a novel four-dimensional Einstein-Gauss-Bonnet (4EGB) theory of gravity was proposed by Glavan and Lin [D. Glavan and C. Lin, Phys. Rev. Lett. 124, 081301 (2020)] which includes a regularized Gauss-Bonnet term by using the re-scaling of the Gauss-Bonnet coupling constant $α\to α/(D-4)$ in the limit $D\to 4$. This theory also has been reformulated to a specific class of the Horndeski theory with an additional scalar degree of freedom and to a spatial covariant version with a Lagrangian multiplier that can eliminate the scalar mode. Here we study the physical properties of the electromagnetic radiation emitted from a thin accretion disk around the static spherically symmetric black hole in the 4EGB gravity. For this purpose, we assume the disk is in a steady-state and in hydrodynamic and thermodynamic equilibrium so that the emitted electromagnetic radiation is a black body spectrum. We study in detail the effects of the Gauss-Bonnet coupling constant $α$ in 4EGB gravity on the energy flux, temperature distribution, and electromagnetic spectrum of the disk. It is shown that with the increases of the parameter $α$, the energy flux, temperature distribution, and electromagnetic spectrum of the accretion disk all increase. Besides, we also show that the accretion efficiency increases as the growth of the parameter $α$. Our results indicate that the thin accretion disk around the static spherically symmetric black hole in the 4EGB gravity is hotter, more luminosity, and more efficient than that around a Schwarzschild black hole with the same mass for a positive $α$, while it is cooler, less luminosity, and less efficient for a negative $α$.

gr-qc↗

arXiv2020

Spherical Accretion Flow onto General Parameterized Spherically Symmetric Black Hole Spacetimes

The transonic phenomenon of black hole accretion and the existence of the photon sphere are the characteristics of strong gravitational fields near a black hole horizon. In this work, we study spherical accretion flow onto a general parametrized spherically symmetric black hole spacetimes. For this purpose, we analyze the accretion process of various perfect fluids, such as the isothermal fluid of ultra-stiff, ultra-relativistic, and sub-relativistic types and polytropic fluid, respectively. The influences of extra parameters beyond the Schwarzschild black hole in the general parameterized spherically symmetric black hole on the flow behaviors of the above-mentioned test fluids are studied in detail. In addition, by studying the accretion of ideal photon gas, we further discuss the correspondence between the sonic radius of accreting photon gas and the photon sphere for the general parameterized spherically symmetric black hole. Some possible future extensions of our analysis are also discussed.

gr-qc↗

arXiv2020

Quasinormal modes, quasiperiodic oscillations and shadow of rotating regular black holes in non-minimally coupled Einstein-Yang-Mills theory

In this paper we obtain an effective metric describing a regular and rotating magnetic black hole (BH) solution with a Yang-Mills electromagnetic source in Einstein-Yang-Mills (EYM) theory using the Newman--Janis algorithm via the non-complexification radial coordinate procedure. We then study the BH shadow and the quasinormal modes (QNMs) for massless scalar and electromagnetic fields and the quasiperiodic oscillations (QPOs). To this end, we also study the embedding diagram for the rotating EYM BH. The energy conditions, shadow curvature radius, topology and the dynamical evolution of scalar and electromagnetic perturbations using the time domain integration method are investigated. We show that the shadow radius decreases by increasing the magnetic charge, while the real part of QNMs of scalar and electromagnetic fields increases by increasing the magnetic charge. This result is consistent with the inverse relation between the shadow radius and the real part of QNMs. In addition, we have studied observational constraints on the EYM parameter $λ$ via frequency analysis of QPOs and the EHT data of shadow cast by the M87 central black hole. We also find that the decaying rate of the EYM BH is slower than that of the neutral and ends up with a tail. We argue that the rotating EYM black hole can be distinguished from the Kerr-Newman black hole with a magnetic charge based on the difference between the angular diameters of their shadows.

gr-qc↗

arXiv2020

Orbital mechanics and quasiperiodic oscillation resonances of black holes in Einstein-Æther theory

In this paper, we study the motion of test particles around two exact charged black-hole solutions in Einstein-Æther theory. Specifically, we first consider the quasi-periodic oscillations (QPOs) and their resonances generated by the particle moving in the Einstein-Æther black hole and then turn to study the periodic orbits of the massive particles. For QPOs, we drop the usually adopted assumptions $ν_U=ν_θ$, $ν_L=ν_r$, and $ν_U/ν_L=3/2$ with $ν_U$ ($ν_L$) and $ν_r$ ($ν_θ$) being the upper (lower) frequency of QPOs and radial (vertical) epicyclic frequency of the orbiting particles, respectively. Instead, we put-forward a new working ansatz for which the Keplerian radius is much closer to that of the innermost stable circular orbit and explore in detail the effects of the æther field on the frequencies of QPOs. We then realize good curves for the frequencies of QPOs, which fit to data of three microquasars very well by ignoring any effects of rotation and magnetic fields. The innermost stable circular orbits (isco) of timelike particles are also analyzed and we find the isco radius increases with increasing $c_{13}$ for the first type black hole while decreases with increasing $c_{14}$ for the second one. We also obtain several periodic orbits and find that they share similar taxonomy schemes as the periodic equatorial orbits in the Schwarzschild/Kerr metrics, in addition to exact solutions for certain choices of the Einstein-Æther parameters. The equations for null geodesics are also briefly considered, where we study circular photon orbits and bending angles for gravitational lensing.

gr-qc↗