arXiv subjects
Publications and source records attributed to Zhiyuan Li.
This paper deals with the unique continuation of solutions for a one-dimensional anomalous diffusion equation with Caputo derivative of order $α\in(0,1)$. Firstly, the uniqueness of solutions to a lateral Cauchy problem for the anomalous diffusion equation is given via the Theta function method, from which we further verify the unique continuation principle.
In this article, for an advection-diffusion equation we study an inverse problem for restoration of source temperature from the information of final temperature profile. The uniqueness of this inverse problem is established by taking an integral transform and using Liouville's theorem (complex analysis).
Despite existing work on ensuring generalization of neural networks in terms of scale sensitive complexity measures, such as norms, margin and sharpness, these complexity measures do not offer an explanation of why neural networks generalize better with over-parametrization. In this work we suggest a novel complexity measure based on unit-wise capacities resulting in a tighter generalization bound for two layer ReLU networks. Our capacity bound correlates with the behavior of test error with increasing network sizes, and could potentially explain the improvement in generalization with over-parametrization. We further present a matching lower bound for the Rademacher complexity that improves over previous capacity lower bounds for neural networks.
Planetary nebulae (PNe) are an important tool for studying the dynamics and chemical evolution of galaxies in the Local Universe, given their characteristic, bright emission line spectra. The Andromeda Galaxy (M31) provides a unique laboratory for studying PNe in the circumnuclear region, thanks to its proximity and almost uniformly low line-of-sight extinction that ensures observations with high resolution and sensitivity. Using the WIYN/Hydra multi-fiber spectrograph, we have obtained optical (4119--6882Å) spectra of 77 PN candidates selected from Hubble Space Telescope narrow-band imaging, which are located within the central $\sim$500pc region of M31. Among these candidates, 49 (64%) are spectroscopically observed for the first time. The spectra of 300 previously known PNe and H II regions, which primarily reside in the disk, are also taken for comparison. All 77 circumnuclear PN candidates exhibit prominent emission lines, including [O III] $λ$5007, [N II] $λ$6583, H$α$ and H$β$, strongly suggesting that they are genuine PNe. We measured the line fluxes, radial velocities and line widths for all objects, and found that the radial velocities of the circumnuclear PNe generally trace rotation of the inner bulge. We also estimated a dynamical mass of $\sim$6.4$\pm$0.5$\times$10$^{9}$$M_{\odot}$ enclosed within an effective galactocentric radius of 340pc, which is compatible with the previously estimated total stellar mass in this region.
We present a study of X-ray source populations in M87, the cD galaxy of the Virgo cluster, using 12 archival {\it Chandra} observations with a total exposure of $\sim$680 ks spanning about a decade. A total of 346 point-like sources are detected down to a limiting 0.5--8 keV luminosity of $4\times10^{37}{\rm~erg~s^{-1}}$ and out to a galactocentric radius of $\sim$40 kpc. We cross-correlate the X-ray sources with published catalogs of globular clusters (GCs), derived from the ACS Virgo Cluster Survey and the Next Generation Virgo Cluster Survey. This results in 122 matches, making it one of the largest samples of GC-hosting X-ray sources in an external galaxy. These sources, most likely low-mass X-ray binaries (LMXBs), correspond to $\sim$5\% of all known GCs within the {\it Chandra} field-of-view. Conversely, $\sim$50\% of the detected X-ray sources are found in a GC. Moreover, red (metal-rich) GCs are $\sim$2.2 times more likely to host an X-ray source than blue (metal-poor) GCs. We also examine 76 currently known ultra-compact dwarf galaxies around M87, but find no significant X-ray counterparts. After statistically accounting for the cosmic X-ray background, we identify $\sim$110 field-LMXBs. The GC-LMXBs and field-LMXBs differ in their luminosity function and radial distribution, which indicates that the latter cannot be primarily originated from GCs. Using another set of deep {\it Chandra} observations toward $\sim$100 kpc northwest of the M87 center, we statistically constrain the abundance of field-LMXBs in the stellar halo, which is consistent with that found in the central region. We also identify 40 variable X-ray sources, among which one source is likely a black hole binary residing in a GC.
We present the study of stellar populations in the central 5.5' (~1.2 kpc) of the M31 bulge by using the optical color magnitude diagram derived from HST ACS WFC/HRC observations. In order to enhance image quality and then obtain deeper photometry, we construct Nyquist-sampled images and use a deconvolution method to detect sources and measure their photometry. We demonstrate that our method performs better than DOLPHOT in the extremely crowded region. The resolved stars in the M31 bulge have been divided into nine annuli and the color magnitude diagram fitting is performed for each of them. We confirm that the majority of stars (> 70%) in the M31 bulge are indeed very old (>5 Gyr) and metal-rich ([Fe/H] > 0.3). At later times, the star formation rate decreased and then experienced a significant rise around 1 Gyr ago, which pervaded the entire M31 bulge. After that, stars formed at less than 500 Myr ago in the central 130". Through simulation, we find that these intermediate-age stars cannot be the artifacts introduced by the blending effect. Our results suggest that although the majority of the M31 bulge are very old, the secular evolutionary process still continuously builds up the M31 bulge slowly. We compare our star formation history with an older analysis derived from the spectral energy distribution fitting, which suggests that the latter one is still a reasonable tool for the study of stellar populations in remote galaxies.
We revisit the question of reducing online learning to approximate optimization of the offline problem. In this setting, we give two algorithms with near-optimal performance in the full information setting: they guarantee optimal regret and require only poly-logarithmically many calls to the approximation oracle per iteration. Furthermore, these algorithms apply to the more general improper learning problems. In the bandit setting, our algorithm also significantly improves the best previously known oracle complexity while maintaining the same regret.
Deep convolutional networks based methods have brought great breakthrough in images classification, which provides an end-to-end solution for handwritten Chinese character recognition(HCCR) problem through learning discriminative features automatically. Nevertheless, state-of-the-art CNNs appear to incur huge computation cost, and require the storage of a large number of parameters especially in fully connected layers, which is difficult to deploy such networks into alternative hardware device with the limit of computation amount. To solve the storage problem, we propose a novel technique called Global Weighted Arverage Pooling for reducing the parameters in fully connected layer without loss in accuracy. Besides, we implement a cascaded model in single CNN by adding mid output layer to complete recognition as early as possible, which reduces average inference time significantly. Experiments were performed on the ICDAR-2013 offline HCCR dataset, and it is found that the proposed approach only needs 6.9ms for classfying a chracter image on average, and achieves the state-of-the-art accuracy of 97.1% while requiring only 3.3MB for storage.
The X-ray emissivity (i.e., luminosity per unit stellar mass) of globular clusters are an important indicator of their dynamical evolution history. Based on deep archival \textit{Chandra} observations, we report a stacking analysis of 44 globular clusters (GCs) with 0.5-8 keV luminosities $L_{\rm X} \lesssim 10^{35} {\rm~erg~s^{-1}}$ in the M\,31 bulge, which are supposed to be dominated by cataclysmic variables (CVs) and coronally active binaries (ABs). We obtain a significant detection at $5σ$ level in 0.5-8 keV band. The average X-ray luminosity per GC and the average X-ray emissivity are determined to be $5.3 \pm 1.6\times10^{33}\rm~erg~s^{-1}$ and $13.2\pm4.3\times10^{27}\rm~erg~{\rm s^{-1}~M^{-1}_{\odot}}$, respectively. Both of these values are consistent with those of MW GCs. What's more, the measured emissivity of M31 GCs is also consistent with that of the MW field stars. Massive GCs have X-ray luminosities which are marginally higher with less massive ones. Massive GCs also show a lower emissivity ($4.5\pm 2.4\times10^{27}{\rm~erg~s^{-1}~M^{-1}_{\odot}}$) than less massive ones ($15.0\pm 7.8\times10^{27}{\rm~erg~s^{-1}~M^{-1}_{\odot}}$), which is consistent with the scenario that the (progenitors of) CVs and ABs were more efficiently destructed via stellar encounters in the more massive GCs. No dependence of the X-ray emissivity on GC color or on the projected galactocentric distance of GCs were found.
Based on archival {\it Chandra} data, we have carried out an X-ray survey of 69, or nearly half the known population of, Milky Way globular clusters (GCs), focusing on weak X-ray sources, mainly cataclysmic variables (CVs) and coronally active binaries (ABs). Using the cumulative X-ray luminosity per unit stellar mass (i.e., X-ray emissivity) as a proxy of the source abundance, we demonstrate a paucity (lower by $41\%\pm27\%$ on average) of weak X-ray sources in most GCs relative to the field, which is represented by the Solar neighborhood and Local Group dwarf elliptical galaxies. We also revisit the mutual correlations among the cumulative X-ray luminosity ($L_X$), cluster mass ($M$) and stellar encounter rate ($Γ$), finding $L_{X}\propto M^{0.74\pm 0.13}$, $L_{X}\proptoΓ^{0.67\pm0.07}$ and $Γ\propto M^{1.28 \pm 0.17}$. The three quantities can further be expressed as $L_{X} \propto M^{0.64\pm0.12}\ Γ^{0.19 \pm0.07}$, which indicates that the dynamical formation of CVs and ABs through stellar encounters in GCs is less dominant than previously suggested, and that the primordial formation channel has a substantial contribution. Taking these aspects together, we suggest that a large fraction of primordial, soft binaries have been disrupted in binary-single or binary-binary stellar interactions before they can otherwise evolve into X-ray-emitting close binaries, whereas the same interactions also have led to the formation of new close binaries. No significant correlations between $L_{X}/L_{K}$ and cluster properties, including dynamical age, metallicity and structural parameters, are found.
In this paper, we discuss an initial-boundary value problem (IBVP) for the multi-term time-fractional diffusion equation with x-dependent coefficients. By means of the Mittag-Leffler functions and the eigenfunction expansion, we reduce the IBVP to an equivalent integral equation to show the unique existence and the analyticity of the solution for the equation. Especially, in the case where all the coefficients of the time-fractional derivatives are non-negative, by the Laplace and inversion Laplace transforms, it turns out that the decay rate of the solution for long time is dominated by the lowest order of the time-fractional derivatives. Finally, as an application of the analyticity of the solution, the uniqueness of an inverse problem in determining the fractional orders in the multi-term time-fractional diffusion equations from one interior point observation is established.
We present an updated catalog of X-ray point sources in the inner 500$\arcsec$ ($\sim$20 parsec) of the Galactic Center (GC), where the {\it nuclear star cluster} (NSC) stands, based on a total of $\sim$4.5 Ms of {\it Chandra} observations taken from September 1999 to April 2013. This ultra-deep dataset offers unprecedented sensitivity for detecting X-ray sources in the GC, down to an intrinsic 2--10 keV luminosity of $1.0\times10^{31}{\rm~erg~s^{-1}}$. A total of 3619 sources are detected in the 2--8 keV band, among which $\sim$3500 are probable GC sources and $\sim$1300 are new identifications. The GC sources collectively account for $\sim$20\% of the total 2--8 keV flux from the inner 250$\arcsec$ region where detection sensitivity is the greatest. Taking advantage of this unprecedented sample of faint X-ray sources that primarily traces the old stellar populations in the NSC, we revisit global source properties, including long-term variability, cumulative spectra, luminosity function and spatial distribution. Based on the equivalent width and relative strength of the iron lines, we suggest that in addition to the arguably predominant population of magnetic cataclysmic variables (CVs), non-magnetic CVs contribute substantially to the detected sources, especially in the lower-luminosity group. On the other hand, the X-ray sources have a radial distribution closely following the stellar mass distribution in the NSC, but much flatter than that of the known X-ray transients, which are presumably low-mass X-ray binaries (LMXBs) caught in outburst. This, together with the very modest long-term variability of the detected sources, strongly suggests that quiescent LMXBs are a minor ($\lesssim$ a few percent) population.
In this short note, we study the asymptotic Chow polystability of toric Del Pezzo surfaces appear in the moduli space of Kähler-Einstein Fano varieties constructed in [OSS16].
Recognizing facial action units (AUs) during spontaneous facial displays is a challenging problem. Most recently, Convolutional Neural Networks (CNNs) have shown promise for facial AU recognition, where predefined and fixed convolution filter sizes are employed. In order to achieve the best performance, the optimal filter size is often empirically found by conducting extensive experimental validation. Such a training process suffers from expensive training cost, especially as the network becomes deeper. This paper proposes a novel Optimized Filter Size CNN (OFS-CNN), where the filter sizes and weights of all convolutional layers are learned simultaneously from the training data along with learning convolution filters. Specifically, the filter size is defined as a continuous variable, which is optimized by minimizing the training loss. Experimental results on two AU-coded spontaneous databases have shown that the proposed OFS-CNN is capable of estimating optimal filter size for varying image resolution and outperforms traditional CNNs with the best filter size obtained by exhaustive search. The OFS-CNN also beats the CNN using multiple filter sizes and more importantly, is much more efficient during testing with the proposed forward-backward propagation algorithm.
Over the past few years, Convolutional Neural Networks (CNNs) have shown promise on facial expression recognition. However, the performance degrades dramatically under real-world settings due to variations introduced by subtle facial appearance changes, head pose variations, illumination changes, and occlusions. In this paper, a novel island loss is proposed to enhance the discriminative power of the deeply learned features. Specifically, the IL is designed to reduce the intra-class variations while enlarging the inter-class differences simultaneously. Experimental results on four benchmark expression databases have demonstrated that the CNN with the proposed island loss (IL-CNN) outperforms the baseline CNN models with either traditional softmax loss or the center loss and achieves comparable or better performance compared with the state-of-the-art methods for facial expression recognition.
We examine initial-boundary value problems for diffusion equations with distributed order time-fractional derivatives. We prove existence and uniqueness results for the weak solution to these systems, together with its continuous dependency on initial value and source term. Moreover, under suitable assumption on the source term, we establish that the solution is analytic in time.
We present a survey of X-ray point sources in the nearest and dynamically young galaxy cluster, Virgo, using archival Chandra observations that sample the vicinity of 80 early-type member galaxies. The X-ray source populations at the outskirt of these galaxies are of particular interest. We detect a total of 1046 point sources (excluding galactic nuclei) out to a projected galactocentric radius of $\sim$40 kpc and down to a limiting 0.5-8 keV luminosity of $\sim$$2\times10^{38}{\rm~erg~s^{-1}}$. Based on the cumulative spatial and flux distributions of these sources, we statistically identify $\sim$120 excess sources that are not associated with the main stellar content of the individual galaxies, nor with the cosmic X-ray background. This excess is significant at a 3.5 $σ$ level, when Poisson error and cosmic variance are taken into account. On the other hand, no significant excess sources are found at the outskirt of a control sample of field galaxies, suggesting that at least some fraction of the excess sources around the Virgo galaxies are truly intra-cluster X-ray sources. Assisted with ground-based and HST optical imaging of Virgo, we discuss the origins of these intra-cluster X-ray sources, in terms of supernova-kicked low-mass X-ray binaries (LMXBs), globular clusters, LMXBs associated with the diffuse intra-cluster light, stripped nucleated dwarf galaxies and free-floating massive black holes.
The Andromeda galaxy (M\,31) hosts one of the nearest and most quiescent super-massive black holes, which provides a rare, but promising opportunity for studying the physics of black hole accretion at the lowest state. We have conducted a multi-frequency, multi-epoch observing campaign, using the Karl G. Jansky Very Large Array (VLA) in its extended configurations in 2011-2012, to advance our knowledge of the still poorly known radio properties of M\,31*. For the first time, we detect M\,31* at 10, 15 and 20 GHz, and measure its spectral index, $α\approx -0.45\pm0.08$ (S$_ν$ $\varpropto$ $ν^α$), over the frequency range of 5-20 GHz. The relatively steep spectrum suggests that the observed radio flux is dominated by the optically-thin part of a putative jet, which is located at no more than a few thousand Schwarzschild radii from the black hole. On the other hand, our sensitive radio images show little evidence for an extended component, perhaps except for several parsec-scale "plumes", the nature of which remains unclear. Our data also reveal significant (a few tens of percent) flux variation of M\,31* at 6 GHz, on timescales of hours to days. Furthermore, a curious decrease of the mean flux density, by $\sim$50\%, is found between VLA observations taken during 2002-2005 and our new observations, which might be associated with a substantial increase in the mean X-ray flux of M\,31* starting in 2006.