arXiv subjects

Xinyu Dai

Publications and source records attributed to Xinyu Dai.

At least 109 records · Page 6Linked to original sources

arXiv2020

R3: A Reading Comprehension Benchmark Requiring Reasoning Processes

Existing question answering systems can only predict answers without explicit reasoning processes, which hinder their explainability and make us overestimate their ability of understanding and reasoning over natural language. In this work, we propose a novel task of reading comprehension, in which a model is required to provide final answers and reasoning processes. To this end, we introduce a formalism for reasoning over unstructured text, namely Text Reasoning Meaning Representation (TRMR). TRMR consists of three phrases, which is expressive enough to characterize the reasoning process to answer reading comprehension questions. We develop an annotation platform to facilitate TRMR's annotation, and release the R3 dataset, a \textbf{R}eading comprehension benchmark \textbf{R}equiring \textbf{R}easoning processes. R3 contains over 60K pairs of question-answer pairs and their TRMRs. Our dataset is available at: \url{http://anonymous}.

cs.CL↗

arXiv2020

Confirmed short periodic variability of subparsec supermassive binary black hole candidate Mrk 231

Here we confirm the short periodic variability of a subparsec supermassive binary black hole (SMBBH) candidate Mrk 231 in the extended optical photometric data set collected by the Catalina Real-Time Transient Survey (CRTS) and All-Sky Automated Survey for Supernovae (ASAS-SN). Using the Lomb-Scargle periodogram and 2DHybrid method, we detected the significant periodicity of ~ 1.1 yr beyond a damped random walk model in the CRTS+ASAS-SN optical data set. Mrk 231 has been previously proposed as a SMBBH candidate with a highly unequal mass ratio (q~ 0.03), very tight mutual separation of ~590 AU, and an orbital period of ~1.2 yr. Hence, our result further supports, even though not prove, the intriguing hypothesis that SMBBHs with low mass ratios may be more common than close-equal mass SMBBHs. This result, however, was obtained from the contribution of CRTS data with limited sampling cadence and photometric accuracy, and further monitoring of Mrk 231 is crucial to confirm the periodicity.

astro-ph.GA↗

arXiv2019

Deficit of Luminous and Normal Red Galaxies in Cosmic Voids

We construct a sample of 10,680 wall galaxies and 3,064 void galaxies with $M_R \lesssim -20$ by cross referencing a void catalog from literature with Baryon Oscillation Spectroscopic Survey (BOSS) CMASS and WiggleZ galaxies, where the CMASS survey targets redder galaxies and the WiggleZ survey targets bluer galaxies. Comparing the density profiles of the red and blue galaxies as a function of the void radius, we find that the number ratio of red-to-blue galaxies increases with distances from the void centers, suggesting a deficit of luminous and normal red galaxies in voids. We find a mean (g -- r) magnitude color of 1.298 and 1.210 for the wall and void galaxies, respectively, when considering the combined red and blue sample, which is found to be a significant difference. However, when considering the blue and red samples separately, we find no significant color difference. We conclude that the constituents galaxies of each population, rather than intrinsic color difference, is the main driver in the apparent average color difference of galaxies in voids and walls, indicating a deficit of luminous and normal red galaxies in voids. Our analysis suggests that the primary environmental-dependence effect on galaxy evolution for normal and luminous galaxies between void and wall regions is manifested in the number of red galaxies, which depends on the environmental-dependent merger history. Using a semi-analytic simulation model, we can successfully reproduce the apparent color difference between the void and wall galaxies.

astro-ph.GA↗

arXiv2019

Confirmation of Planet-Mass Objects in Extragalactic Systems

Quasar microlensing serves as a unique probe of discrete objects within galaxies and galaxy clusters. Recent advancement of the technique shows that it can constrain planet-scale objects beyond our native galaxy by studying their induced microlensing signatures, the energy shift of emission lines originated in the vicinity of the black hole of high redshift background quasars. We employ this technique to exert effective constraints on the planet-mass object distribution within two additional lens systems, Q J0158$-$4325 ($z_l = 0.317$) and SDSS J1004+4112 ($z_l = 0.68$) using Chandra observations of the two gravitationally-lensed quasars. The observed variations of the emission line peak energy can be explained as microlensing of the FeK$α$ emission region induced by planet-mass microlenses. To corroborate this, we perform microlensing simulations to determine the probability of a caustic transiting the source region and compare this with the observed line shift rates. Our analysis yields constraints on the sub-stellar population, with masses ranging from Moon ($10^{-8} M_{\odot}$) to Jupiter ($10^{-3} M_{\odot}$) sized bodies, within these galaxy or cluster scale structures, with total mass fractions of $\sim 3\times10^{-4}$ and $\sim 1\times10^{-4}$ with respect to halo mass for Q J0158$-$4325 and SDSS J1004+4112, respectively. Our analysis suggests that unbound planet-mass objects are universal in galaxies, and we surmise the objects to be either free-floating planets or primordial black holes. We present the first-ever constraints on the sub-stellar mass distribution in the intra-cluster light of a galaxy cluster. Our analysis yields the most stringent limit for primordial black holes at the mass range.

astro-ph.GA↗

arXiv2019

X-ray monitoring of gravitationally lensed radio-loud quasars with Chandra

In this work, we calculated the sizes of unresolved X-ray emission regions in three gravitationally lensed radio-loud quasars, B\,1422+231, MG\,J0414+0534 and Q\,0957+561, using a combination of imaging and spectral analysis on the X-ray data taken from the \textit{Chandra X-Ray Observatory}. We tentatively detected FeK$α$ emission lines in MG\,J0414+0534 and Q\,0957+561 with over 95\% significance, whereas, we did not significantly detect FeK$α$ emission in B\,1422+231. We constructed differential microlensing light curves from absorption corrected count rates. We subsequently performed a microlensing analysis on the X-ray microlensing light curves to measure the X-ray source sizes in soft (0.83--3.6 keV), hard (3.6--21.8 keV), and full (0.83--21.8 keV) bands, based on either Bayesian or maximum likelihood probabilities. For B\,1422+231, sizes from the two methods are consistent with each other, e.g. $R_X^{hard}/R_G = 6.17\pm5.48 \text{ (Bayesian), } 11.81\pm3.75 \text{ (maximum likelihood)}$, where $R_G=GM_{BH}/c^2)$. However, for MG\,J0414+0534 and Q\,0957+561, the two methods yield completely different results suggesting that more frequently sampled data with better signal-to-noise ratio are needed to measure the source size for these two objects. Comparing the acquired size values with the radio-quiet sample in the literature we found that our results are consistent with X-ray source size scaling approximately as $R_X \propto M_{BH}$ with the mass of the central supermassive black hole. Our results also indicate that radio-loud quasars tend to have larger unresolved X-ray emission sizes compared to the radio-quiet ones.

astro-ph.CO↗

arXiv2019

Fine-grained Knowledge Fusion for Sequence Labeling Domain Adaptation

In sequence labeling, previous domain adaptation methods focus on the adaptation from the source domain to the entire target domain without considering the diversity of individual target domain samples, which may lead to negative transfer results for certain samples. Besides, an important characteristic of sequence labeling tasks is that different elements within a given sample may also have diverse domain relevance, which requires further consideration. To take the multi-level domain relevance discrepancy into account, in this paper, we propose a fine-grained knowledge fusion model with the domain relevance modeling scheme to control the balance between learning from the target domain data and learning from the source domain model. Experiments on three sequence labeling tasks show that our fine-grained knowledge fusion model outperforms strong baselines and other state-of-the-art sequence labeling domain adaptation methods.

cs.CL↗

arXiv2019

GeoSQA: A Benchmark for Scenario-based Question Answering in the Geography Domain at High School Level

Scenario-based question answering (SQA) has attracted increasing research attention. It typically requires retrieving and integrating knowledge from multiple sources, and applying general knowledge to a specific case described by a scenario. SQA widely exists in the medical, geography, and legal domains---both in practice and in the exams. In this paper, we introduce the GeoSQA dataset. It consists of 1,981 scenarios and 4,110 multiple-choice questions in the geography domain at high school level, where diagrams (e.g., maps, charts) have been manually annotated with natural language descriptions to benefit NLP research. Benchmark results on a variety of state-of-the-art methods for question answering, textual entailment, and reading comprehension demonstrate the unique challenges presented by SQA for future research.

cs.CL↗

arXiv2019

Generating Sentences from Disentangled Syntactic and Semantic Spaces

Variational auto-encoders (VAEs) are widely used in natural language generation due to the regularization of the latent space. However, generating sentences from the continuous latent space does not explicitly model the syntactic information. In this paper, we propose to generate sentences from disentangled syntactic and semantic spaces. Our proposed method explicitly models syntactic information in the VAE's latent space by using the linearized tree sequence, leading to better performance of language generation. Additionally, the advantage of sampling in the disentangled syntactic and semantic latent spaces enables us to perform novel applications, such as the unsupervised paraphrase generation and syntax-transfer generation. Experimental results show that our proposed model achieves similar or better performance in various tasks, compared with state-of-the-art related work.

cs.CL↗

arXiv2019

Adversarial Sub-sequence for Text Generation

Generative adversarial nets (GAN) has been successfully introduced for generating text to alleviate the exposure bias. However, discriminators in these models only evaluate the entire sequence, which causes feedback sparsity and mode collapse. To tackle these problems, we propose a novel mechanism. It first segments the entire sequence into several sub-sequences. Then these sub-sequences, together with the entire sequence, are evaluated individually by the discriminator. At last these feedback signals are all used to guide the learning of GAN. This mechanism learns the generation of both the entire sequence and the sub-sequences simultaneously. Learning to generate sub-sequences is easy and is helpful in generating an entire sequence. It is easy to improve the existing GAN-based models with this mechanism. We rebuild three previous well-designed models with our mechanism, and the experimental results on benchmark data show these models are improved significantly, the best one outperforms the state-of-the-art model.\footnote[1]{All code and data are available at https://github.com/liyzcj/seggan.git

cs.LG↗

arXiv2019

Constraining Quasar Relativistic Reflection Regions and Spins with Microlensing

We present an analysis of Chandra spectra of five gravitationally lensed active galactic nuclei. We confirm the previous detections of FeK$α$ emission lines in most images of these objects with high significance. The line energies range from 5.8 to 6.8 keV with widths from unresolved to 0.6 keV, consistent with emission close to spinning black holes viewed at different inclination angles. We also confirm the positive offset from the Iwasawa-Taniguchi effect, the inverse correlation between the FeK$α$ equivalent width and the X-ray luminosity in AGN, where our measured equivalent widths are larger in lensed quasars. We attribute this effect to microlensing, and perform a microlensing likelihood analysis to constrain the emission size of the relativistic reflection region and the spin of supermassive black holes, assuming that the X-ray corona and the reflection region, responsible for the iron emission line, both follow power-law emissivity profiles. The microlensing analysis yields strong constraints on the spin and emissivity index of the reflection component for Q2237+0305, with $a > 0.92$ and $n > 5.4$. For the remaining four targets, we jointly constrain the two parameters, yielding $a=0.8\pm0.16$ and an emissivity index of $n=4.0\pm 0.8$, suggesting that the relativistic X-ray reflection region is ultra-compact and very close to the innermost stable circular orbits of black holes, which are spinning at close to the maximal value. We successfully constrain the half light radius of the emission region to $< 2.4$ $r_g$ ($r_g = GM/c^2$) for Q2237+0305 and in the range 5.9--7.4 $r_g$ for the joint sample.

astro-ph.HE↗

arXiv2019

Astro2020 Science White Paper - Quasar Microlensing: Revolutionizing our Understanding of Quasar Structure and Dynamics

Microlensing by stars within distant galaxies acting as strong gravitational lenses of multiply-imaged quasars, provides a unique and direct measurement of the internal structure of the lensed quasar on nano-arcsecond scales. The measurement relies on the temporal variation of high-magnification caustic crossings which vary on timescales of days to years. Multiwavelength observations provide information from distinct emission regions in the quasar. Through monitoring of these strong gravitational lenses, a full tomographic view can emerge with Astronomical-Unit scale resolution. Work to date has demonstrated the potential of this technique in about a dozen systems. In the 2020s there will be orders of magnitude more systems to work with. Monitoring of lens systems for caustic-crossing events to enable triggering of multi-platform, multi-wavelength observations in the 2020s will fulfill the potential of quasar microlensing as a unique and comprehensive probe of active black hole structure and dynamics.

astro-ph.GA↗

arXiv2019

The Most Powerful Lenses in the Universe: Quasar Microlensing as a Probe of the Lensing Galaxy

Optical and X-ray observations of strongly gravitationally lensed quasars (especially when four separate images of the quasar are produced) determine not only the amount of matter in the lensing galaxy but also how much is in a smooth component and how much is composed of compact masses (e.g., stars, stellar remnants, primordial black holes, CDM sub-halos, and planets). Future optical surveys will discover hundreds to thousands of quadruply lensed quasars, and sensitive X-ray observations will unambiguously determine the ratio of smooth to clumpy matter at specific locations in the lensing galaxies and calibrate the stellar mass fundamental plane, providing a determination of the stellar $M/L$. A modest observing program with a sensitive, sub-arcsecond X-ray imager, combined with the planned optical observations, can make those determinations for a large number (hundreds) of the lensing galaxies, which will span a redshift range of $\sim$$0.25<z<1.5$

astro-ph.GA↗

arXiv2018

Reconciling the Quasar Microlensing Disc Size Problem with a Wind Model of Active Galactic Nucleus

Many analyses have concluded that the accretion disc sizes measured from the microlensing variability of quasars are larger than the expectations from the standard thin disc theory by a factor of $\sim4$. We propose a simply model by invoking a strong wind from the disc to flatten its radial temperature profile, which can then reconcile the size discrepancy problem. This wind model has been successfully applied to several microlensed quasars with a wind strength $s\lesssim1.3$ by only considering the inward decreasing of the mass accretion rate (where $s$ is defined through $\dot{M}(R)\propto({R}/{R_{0}})^{s}$ ). After further incorporating the angular momentum transferred by the wind, our model can resolve the disc size problem with an even lower wind parameter. The corrected disc sizes under the wind model are correlated with black hole masses with a slope in agreement with our modified thin disc model.

astro-ph.HE↗

arXiv2018

Swift Monitoring Observations of Mrk 231: Detection of Ultraviolet Variability

We analyze 168 Swift monitoring observations of the nearest broad absorption line quasar Mrk 231 in the UV and X-ray bands, where we detect significant variability in the UV ($\sim$2246Å) light curve with a null probability of $4.3\times10^{-10}$ for a constant model. Separately, from an archival sample of Swift observed active galactic nuclei (AGN), we measure the relation between UV excess variance and luminosity, finding that the normalized UV excess variance decreases with luminosity. Comparing to this mean relation, the normalized UV excess variance of Mrk 231 is smaller, however within the scatter characterising the full population. The upper limit of the X-ray excess variance is consistent with other AGN. The power spectrum density of the UV light curve can be well fit by a power law model with a slope of $1.82\pm0.14$ between $10^{-7.5}$ and $10^{-6}$ Hz, consistent with those for typical AGN, with no obvious quasi-periodical oscillation peaks. The UV variability and its power spectrum suggest that a significant amount of the UV emission of Mrk 231 is from the accretion disk. The consistencies in the normalized UV variability and the shape of the power spectrum density between Mrk 231 and other normal AGN suggest that the origin of UV variability of broad absorption line quasars is similar to other AGN, and dust scattering at large scales such as the torus is not a dominating process for the UV emission of Mrk 231. Significant scattering, if present, is constrained to smaller than $\sim$10 light days. We perform lagged correlation analysis between the UV and X-ray light curves and find the correlation insignificant within the present data.

astro-ph.GA↗

arXiv2018

A Second-Order Moment of Microlensing Variability as a Novel Tool in Extragalactic Research

We define a second-order moment of the observational differential microlensing curves that can be used to impose constraints on physical properties of lensed quasars. We show that this quantity is sensitive both to variations in the source size and the deflector mass. We formulize a methodology to recover the source size from the observational measurements when the mass spectrum is fixed. As a case study, we test it with a sample of four quadruple lenses, both in simulated scenarios and with real data from the $\textit{Chandra X-ray Observatory}$. In our simulations with a uniform stellar population the method works best to detect sources around $0.1$ Einstein radii, giving correct upper/lower limits for much smaller/bigger sizes without requiring a big leap in additional computational effort as compared to a single-epoch approach, yet taking advantage of multi-epoch information. We apply the method to a small sample of X-ray data from four objects assuming a range of star masses, and obtain a degeneracy relation between the source size and deflector mass. Combined with previous estimates for the size of the X-ray corona, the degeneracy relation suggests that X-ray microlensing is mainly induced by planetary mass objects.

astro-ph.GA↗

arXiv2018

The Extended Distribution of Baryons Around Galaxies

We summarize and reanalyze observations bearing upon missing galactic baryons, where we propose a consistent picture for halo gas in L >~ L* galaxies. The hot X-ray emitting halos are detected to 50-70 kpc, where typically, M_hot(<50 kpc) ~ 5E9 Msun, and with density n \propto r^-3/2. When extrapolated to R200, the gas mass is comparable to the stellar mass, but about half of the baryons are still missing from the hot phase. If extrapolated to 1.9-3 R200, the baryon to dark matter ratio approaches the cosmic value. Significantly flatter density profiles are unlikely for R < 50 kpc and they are disfavored but not ruled out for R > 50 kpc. For the Milky Way, the hot halo metallicity lies in the range 0.3-1 solar for R < 50 kpc. Planck measurements of the thermal Sunyaev-Zeldovich effect toward stacked luminous galaxies (primarily early-type) indicate that most of their baryons are hot, near the virial temperature, and extend beyond R200. This stacked SZ signal is nearly an order of magnitude larger than that inferred from the X-ray observations of individual (mostly spiral) galaxies with M_* > 10^11.3 Msun. This difference suggests that the hot halo properties are distinct for early and late type galaxies, possibly due to different evolutionary histories. For the cooler gas detected in UV absorption line studies, we argue that there are two absorption populations: extended halos; and disks extending to ~50 kpc, containing most of this gas, and with masses a few times lower than the stellar masses. Such extended disks are also seen in 21 cm HI observations and in simulations.

astro-ph.GA↗

arXiv2018

Probing Planets in Extragalactic Galaxies Using Quasar Microlensing

Previously, planets have been detected only in the Milky Way galaxy. Here, we show that quasar microlensing provides a means to probe extragalactic planets in the lens galaxy, by studying the microlensing properties of emission close to the event horizon of the supermassive black hole of the background quasar, using the current generation telescopes. We show that a population of unbound planets between stars with masses ranging from Moon to Jupiter masses is needed to explain the frequent Fek line energy shifts observed in the gravitationally lensed quasar RXJ1131-1231 at a lens redshift of $z=0.295$ or 3.8 billion light-years away. We constrain the planet mass fraction to be larger than 0.0001 of the halo mass, which is equivalent to 2,000 objects ranging from Moon to Jupiter mass per main sequence star.

astro-ph.GA↗

arXiv2017

Modeling Past and Future for Neural Machine Translation

Existing neural machine translation systems do not explicitly model what has been translated and what has not during the decoding phase. To address this problem, we propose a novel mechanism that separates the source information into two parts: translated Past contents and untranslated Future contents, which are modeled by two additional recurrent layers. The Past and Future contents are fed to both the attention model and the decoder states, which offers NMT systems the knowledge of translated and untranslated contents. Experimental results show that the proposed approach significantly improves translation performance in Chinese-English, German-English and English-German translation tasks. Specifically, the proposed model outperforms the conventional coverage model in both of the translation quality and the alignment error rate.

cs.CL↗