arXiv subjects

Zhiyuan Li

Publications and source records attributed to Zhiyuan Li.

At least 361 records · Page 20Linked to original sources

arXiv2019

Unique determination of several coefficients in a fractional diffusion(-wave) equation by a single measurement

We consider the inverse problem of determining different type of information about a diffusion process, described by ordinary or fractional diffusion equations stated on a bounded domain, like the density of the medium or the velocity field associated with the moving quantities from a single boundary measurement. This properties will be associated with some general class of time independent coefficients that we recover from a single Neumann boundary measurement, on some parts of the boundary, of the solution of our diffusion equation with a suitable boundary input, located on some parts of the boundary.

math.AP↗

arXiv2019

Feature-level and Model-level Audiovisual Fusion for Emotion Recognition in the Wild

Emotion recognition plays an important role in human-computer interaction (HCI) and has been extensively studied for decades. Although tremendous improvements have been achieved for posed expressions, recognizing human emotions in "close-to-real-world" environments remains a challenge. In this paper, we proposed two strategies to fuse information extracted from different modalities, i.e., audio and visual. Specifically, we utilized LBP-TOP, an ensemble of CNNs, and a bi-directional LSTM (BLSTM) to extract features from the visual channel, and the OpenSmile toolkit to extract features from the audio channel. Two kinds of fusion methods, i,e., feature-level fusion and model-level fusion, were developed to utilize the information extracted from the two channels. Experimental results on the EmotiW2018 AFEW dataset have shown that the proposed fusion methods outperform the baseline methods significantly and achieve better or at least comparable performance compared with the state-of-the-art methods, where the model-level fusion performs better when one of the channels totally fails.

cs.CV↗

arXiv2019

Experimental Measurement of the Quantum Metric Tensor and Related Topological Phase Transition with a Superconducting Qubit

Berry curvature is an imaginary component of the quantum geometric tensor (QGT) and is well studied in many branches of modern physics; however, the quantum metric as a real component of the QGT is less explored. Here, by using tunable superconducting circuits, we experimentally demonstrate two methods to directly measure the quantum metric tensor for characterizing the geometry and topology of underlying quantum states in parameter space. The first method is to probe the transition probability after a sudden quench, and the second one is to detect the excitation rate under weak periodic driving. Furthermore, based on quantum-metric and Berry-curvature measurements, we explore a topological phase transition in a simulated time-reversal-symmetric system, which is characterized by the Euler characteristic number instead of the Chern number. The work opens up a unique approach to explore the topology of quantum states with the QGT.

quant-ph↗

arXiv2019

Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks

Recent works have cast some light on the mystery of why deep nets fit any data and generalize despite being very overparametrized. This paper analyzes training and generalization for a simple 2-layer ReLU net with random initialization, and provides the following improvements over recent works: (i) Using a tighter characterization of training speed than recent papers, an explanation for why training a neural net with random labels leads to slower training, as originally observed in [Zhang et al. ICLR'17]. (ii) Generalization bound independent of network size, using a data-dependent complexity measure. Our measure distinguishes clearly between random labels and true labels on MNIST and CIFAR, as shown by experiments. Moreover, recent papers require sample complexity to increase (slowly) with the size, while our sample complexity is completely independent of the network size. (iii) Learnability of a broad class of smooth functions by 2-layer ReLU nets trained via gradient descent. The key idea is to track dynamics of training and generalization via properties of a related kernel.

cs.LG↗

arXiv2019

Inverse Problems of Determining Sources of the Fractional Partial Differential Equations

In this chapter, we mainly review theoretical results on inverse source problems for diffusion equations with the Caputo time-fractional derivatives of order $α\in(0,1)$. Our survey covers the following types of inverse problems: 1. determination of time-dependent functions in interior source terms 2. determination of space-dependent functions in interior source terms 3. determination of time-dependent functions appearing in boundary conditions

math.AP↗

arXiv2019

Inverse Problems of Determining Parameters of the Fractional Partial Differential Equations

When considering fractional diffusion equation as model equation in analyzing anomalous diffusion processes, some important parameters in the model related to orders of the fractional derivatives, are often unknown and difficult to be directly measured, which requires one to discuss inverse problems of identifying these physical quantities from some indirectly observed information of solutions. Inverse problems in determining these unknown parameters of the model are not only theoretically interesting, but also necessary for finding solutions to initial-boundary value problems and studying properties of solutions. This chapter surveys works on such inverse problems for fractional diffusion equations.

math.AP↗

arXiv2019

Inverse Problems of Determining Coefficients of the Fractional Partial Differential Equations

When considering fractional diffusion equation as model equation in analyzing anomalous diffusion processes, some important parameters in the model, for example, the orders of the fractional derivative or the source term, are often unknown, which requires one to discuss inverse problems to identify these physical quantities from some additional information that can be observed or measured practically. This chapter investigates several kinds of inverse coefficient problems for the fractional diffusion equation.

math.AP↗

arXiv2019

Exploring the Mass Segregation Effect of X-ray Sources in Globular Clusters: The Case of 47 Tucanae

Using archival {\it Chandra} observations with a total exposure of 510 ks, we present an updated catalog of point sources for Globular Cluster 47 Tucanae. Our study covers an area of $\sim 176.7$ arcmin$^{2}$ (i.e., with $R\lesssim7.5\arcmin$) with 537 X-ray sources. We show that the surface density distribution of X-ray sources in 47 Tuc is highly peaked in cluster center, rapidly decreases at intermediate radii, and finally rises again at larger radii, with two distribution dips at $R\sim 100\arcsec$ and $R\sim 170\arcsec$ for the faint ($L_{X}\lesssim 5.0\times 10^{30} {\rm\ erg\,s^{-1}}$) and bright ($L_{X}\gtrsim 5.0\times 10^{30} {\rm\ erg\,s^{-1}}$) groups of X-ray sources, separately. These distribution features are similar to those of Blue Straggler Stars (BSS), where the distribution dip is located at $R\sim 200\arcsec$ \citep{ferraro2004}. By fitting the radial distribution of each group of sources with a "generalized King model", we estimated an average mass of $1.51\pm0.17\ M_{\odot}$, $1.44\pm0.15\ M_{\odot}$ and $1.16\pm0.06\ M_{\odot}$ for the BSS, bright and faint X-ray sources, respectively. These results are consistent with the mass segregation effect of heavy objects in GCs, where more massive objects drop to the cluster center faster and their distribution dip propagates outward further. Besides, the peculiar distribution profiles of X-ray sources and BSS are also consistent with the mass segregation model of binaries in GCs, which suggests that in addition to the dynamical formation channel, primordial binaries are also a significant contributor to the X-ray source population in GCs.

astro-ph.HE↗

arXiv2019

A Deep Chandra View of A Candidate Parsec-Scale Jet from the Galactic Center Super-massive Black Hole

We have investigated the linear X-ray filament, G359.944-0.052, previously identified as a likely X-ray counterpart of a parsec-scale jet from the Galactic Center super-massive black hole (SMBH), Sagittarius A* (Sgr A*), using a total of ~5.6 Ms ultra-deep Chandra observations taken from September 1999 to July 2017. This unprecedented dataset enables us to examine flux and spectral variations that might be related to intrinsic properties of the weakly accreting SMBH. We find no flux or spectral variation in G359.944-0.052 after the G2 periapsis passage around early 2014, however, a moderate flux increase of ~2 sigma significance might be associated with the periapsis passage of G1 in early 2001. The filament exhibits an unusually hard spectrum (photon-index < 1) in its portion closest to Sgr A* (i.e., near-side) and a significant spectral softening in the more distant portion, which can be interpreted as synchrotron cooling of the relativistic electrons moving along the jet path. In particular, the hard spectrum of the near-side suggests a piling up of quasi-monoenergetic electrons caused by rapid radiative cooling. The spectral and temporal properties of G359.944-0.052 strengthen the case of it being the X-ray counterpart of a jet launched by Sgr A*.

astro-ph.HE↗

arXiv2019

Chandra Detection of Intra-cluster X-ray Sources in Fornax

Based on archival {\it Chandra} observations with a total exposure of 1.3 Ms, we study X-ray point sources in the Fornax cluster of galaxies, with the primary aim of searching for intra-cluster X-ray source populations. We detect 1177 point sources out to a projected radius of $\sim$30 arcmin ($\sim$180 kpc) from the cluster center and down to a limiting 0.5--8 keV luminosity of $\sim3\times10^{37}{\rm~erg~s^{-1}}$. We construct source surface density profile, after excluding sources associated with foreground stars, known globular clusters, ultra-compact dwarfs and galactic nuclei. From this profile we statistically identify $\sim$183 excess sources that are not associated with the bulk stellar content of the individual member galaxies of Fornax, nor with the cosmic X-ray background. Taking into account Poisson error and cosmic variance, the cumulative significance of this excess is at $\gtrsim 2\,σ$ level (with a maximum of 3.6\,$σ$) outside three effective radii of the central giant elliptical, NGC\,1399. The luminosity function of the excess sources is found to be significantly steeper than that of the GC-hosting sources (presumably low-mass X-ray binaries [LMXBs]), disfavoring the possibility that unidentified GCs are primarily responsible for the excess. We show that a large fraction of the excess can be related to the extended stellar halo of NGC\,1399 and/or the diffuse intra-cluster light, thus providing strong evidence for the presence of intra-cluster X-ray sources in Fornax, the second unambiguous case for a galaxy cluster after Virgo. Other possible origins of the excess, including supernova-kicked LMXBs and stripped nucleated dwarf galaxies are discussed.

astro-ph.GA↗

arXiv2018

JCMT Mapping of CO(3-2) in the Circumnuclear Region of M31

We present a survey of CO(3-2) molecular line emission in the circumnuclear region of M31 with the James Clerk Maxwell Telescope (JCMT), aiming to explore the physical conditions of the molecular gas. Significant CO(3-2) lines are detected primarily along the so-called nuclear spiral, out to a projected galactocentric radius of 700 pc at a linear resolution of ~50 pc. We find that the velocity field of the molecular gas is in rough agreement with that of the ionized gas previously determined from optical observations. Utilizing existed CO(2-1) and CO(1-0) measurements in selected regions of the nuclear spiral, we derive characteristic intensity ratios of CO(3-2)/CO(2-1) and CO(3-2)/CO(1-0), which are both close to unity and are significantly higher than the typical intensity ratios in the disk. Such line ratios suggest high kinetic temperatures of the gas, which might be due to strong interstellar shocks prevalent in the circumnuclear region.

astro-ph.GA↗

arXiv2018

Probabilistic Attribute Tree in Convolutional Neural Networks for Facial Expression Recognition

In this paper, we proposed a novel Probabilistic Attribute Tree-CNN (PAT-CNN) to explicitly deal with the large intra-class variations caused by identity-related attributes, e.g., age, race, and gender. Specifically, a novel PAT module with an associated PAT loss was proposed to learn features in a hierarchical tree structure organized according to attributes, where the final features are less affected by the attributes. Then, expression-related features are extracted from leaf nodes. Samples are probabilistically assigned to tree nodes at different levels such that expression-related features can be learned from all samples weighted by probabilities. We further proposed a semi-supervised strategy to learn the PAT-CNN from limited attribute-annotated samples to make the best use of available data. Experimental results on five facial expression datasets have demonstrated that the proposed PAT-CNN outperforms the baseline models by explicitly modeling attributes. More impressively, the PAT-CNN using a single model achieves the best performance for faces in the wild on the SFEW dataset, compared with the state-of-the-art methods using an ensemble of hundreds of CNNs.

cs.CV↗

arXiv2018

Theoretical Analysis of Auto Rate-Tuning by Batch Normalization

Batch Normalization (BN) has become a cornerstone of deep learning across diverse architectures, appearing to help optimization as well as generalization. While the idea makes intuitive sense, theoretical analysis of its effectiveness has been lacking. Here theoretical support is provided for one of its conjectured properties, namely, the ability to allow gradient descent to succeed with less tuning of learning rates. It is shown that even if we fix the learning rate of scale-invariant parameters (e.g., weights of each layer with BN) to a constant (say, $0.3$), gradient descent still approaches a stationary point (i.e., a solution where gradient is zero) in the rate of $T^{-1/2}$ in $T$ iterations, asymptotically matching the best bound for gradient descent with well-tuned learning rates. A similar result with convergence rate $T^{-1/4}$ is also shown for stochastic gradient descent.

cs.LG↗

arXiv2018

Deep Template Matching for Offline Handwritten Chinese Character Recognition

Just like its remarkable achievements in many computer vision tasks, the convolutional neural networks (CNN) provide an end-to-end solution in handwritten Chinese character recognition (HCCR) with great success. However, the process of learning discriminative features for image recognition is difficult in cases where little data is available. In this paper, we propose a novel method for learning siamese neural network which employ a special structure to predict the similarity between handwritten Chinese characters and template images. The optimization of siamese neural network can be treated as a simple binary classification problem. When the training process has been finished, the powerful discriminative features help us to generalize the predictive power not just to new data, but to entirely new classes that never appear in the training set. Experiments performed on the ICDAR-2013 offline HCCR datasets have shown that the proposed method has a very promising generalization ability to the new classes that never appear in the training set.

cs.CV↗

arXiv2018

Searching for GeV gamma-ray emission from the bulge of M31

The three major large-scale, diffuse $γ$-ray structures of the Milky Way are the Galactic disk, a bulge-like GeV excess towards the Galactic center, and the Fermi bubble. Whether such structures can also be present in other normal galaxies remains an open question. M31, as the nearest massive normal galaxy, holds promise for spatially-resolving the $γ$-ray emission. Based on more than 8 years of Fermi-LAT observations, we use (1) disk, (2) bulge, and (3) disk-plus-bulge templates to model the spatial distribution of the $γ$-ray emission from M31. Among these, the disk-plus-bulge template delivers the best-fit, in which the bulge component has a TS value 25.7 and a photon-index of $2.57\pm0.17$, providing strong evidence for a centrally-concentrated gamma-ray emission from M31, that is analogous to the Galactic center excess. The total 0.2-300 GeV gamma-ray luminosity from this bulge component is $(1.16\pm0.14)\times10^{38} erg s^{-1}$, which would require $\sim1.5\times10^5$ millisecond pulsars, if they were the dominant source. We also search for a Fermi bubble-like structure in M31 using the full dataset (pass8), but no significant evidence is found. In addition, a likelihood analysis using only photons with the most accurate reconstructed direction (i.e., PSF3-only data) reveals a 4.8 $σ$ point-like source located at $\sim$ 10kpc to the northwest of the M31 disk, with a luminosity of $(0.97\pm0.27)\times10^{38} erg s^{-1}$ and a photon-index of $2.31\pm0.18$. Lacking of a counterpart on the southeast side of the disk, the relation between this point-like source and a bubble-like structure remains elusive.

astro-ph.HE↗

arXiv2018

A Chandra Survey of Milky Way Globular Clusters II: Testing the Hills-Heggie Law

Binary-single and binary-binary encounters play a pivotal role in the evolution of star clusters, as they may lead to the disruption or hardening of binaries, a novel prediction of the Hills-Heggie law. Based on our recent {\it Chandra} survey of Galactic globular clusters (GCs), we revisit the role of stellar dynamical interactions in GCs, focusing on main-sequence (MS) binary encounters as a potential formation channel of the observed X-ray sources in GCs. We show that the cumulative X-ray luminosity ($L_X$), a proxy of the total number of X-ray-emitting binaries (primarily cataclysmic variables and coronally active binaries) in a given GC, is highly correlated with the MS binary encounter rate ($Γ_{b}$), as $L_X \propto Γ_{b}^{0.77\pm0.11}$. We further test the Hills-Heggie law against the {\it binary hardness ratio}, defined as the relative number of X-ray-emitting hard binaries to MS binaries and approximated by $L_{X}/(L_{K}f_{b})$, with $L_{K}$ being the GC K-band luminosity and $f_b$ the MS binary fraction. We demonstrate that the binary hardness ratio of most GCs is larger than that of the Solar neighborbood stars, and exhibits a positive correlation with the cluster specific encounter rate ($γ$), as $L_{X}/(L_{K}f_{b}) \propto γ^{0.65\pm0.12}$. We also find a strong correlation between the binary hardness ratio and cluster velocity dispersion ($σ$), with $L_{X}/(L_{K}f_{b}) \propto σ^{1.71\pm0.48}$, which is consistent with the Hills-Heggie law. We discuss the role of binary encounters in the context of the Nuclear Star Cluster, arguing that the X-ray-emitting, close binaries detected therein could have been predominatly formed in GCs that later inspiralled to the Galactic center.

astro-ph.HE↗

arXiv2018

Probing the dynamical state, baryon content, and multiphase nature of galaxy clusters with bright background QSOs

We have initiated a programme to study the physical/dynamical state of gas in galaxy clusters and the impact of the cluster environment on gaseous halos of individual galaxies using X-ray imaging and UV absorption line spectroscopy of background QSOs. Here we report results from the analysis Chandra and XMM-Newton archival data of five galaxy clusters with such QSOs, one of which has an archival UV spectrum. We characterize the gravitational masses and dynamical states, as well as the hot intracluster medium (ICM) properties of these clusters. Most clusters are dynamically disturbed clusters based on the X-ray morphology parameters, the X-ray temperature profiles, the large offset between X-ray peak and brightest cluster galaxy (BCG). The baryon contents in the hot ICM and stars of these clusters within $r_{500}$ are lower than the values expected from the gravitational masses, according to the standard cosmology. We also estimate column densities of the hot ICM along the sightlines toward the background QSOs as well as place upper limits on the warm-hot phase for the one sightline with existing UV observations. These column densities, compared with those of the warm and warm-hot ICM to be measured with UV absorption line spectroscopy, will enable us to probe the relationship among various gaseous phases and their connection to the heating/cooling and dynamical processes of the clusters. Furthermore, our analysis of the archival QSO spectrum probing one cluster underscores the need for high quality, targeted UV observations to robustly constrain the 10$^{5-6}$ K gas phase.

astro-ph.GA↗

arXiv2018

An inverse problem for distributed order time-fractional diffusion equations

This paper deals with the distributed order time-fractional diffusion equations with non-homogeneous Dirichlet (Nuemann) boundary condition. We first prove the wellposedness of the weak solution to the initial boundary value problem for the distributed order time-fractional diffusion equation by means of eigenfunction expansion, which ensure that the weak solution has the classical derivatives. We next give a Harnack type inequality of the solution in the frequency domain under the Laplace transform, from which we further show a uniqueness result for an inverse problem in determining the weight function in the distributed order time derivative from point observation.

math.AP↗