arXiv subjects

Yijun Wang

Publications and source records attributed to Yijun Wang.

At least 91 records · Page 5Linked to original sources

arXiv2023

EMoG: Synthesizing Emotive Co-speech 3D Gesture with Diffusion Model

Although previous co-speech gesture generation methods are able to synthesize motions in line with speech content, it is still not enough to handle diverse and complicated motion distribution. The key challenges are: 1) the one-to-many nature between the speech content and gestures; 2) the correlation modeling between the body joints. In this paper, we present a novel framework (EMoG) to tackle the above challenges with denoising diffusion models: 1) To alleviate the one-to-many problem, we incorporate emotion clues to guide the generation process, making the generation much easier; 2) To model joint correlation, we propose to decompose the difficult gesture generation into two sub-problems: joint correlation modeling and temporal dynamics modeling. Then, the two sub-problems are explicitly tackled with our proposed Joint Correlation-aware transFormer (JCFormer). Through extensive evaluations, we demonstrate that our proposed method surpasses previous state-of-the-art approaches, offering substantial superiority in gesture synthesis.

cs.CV↗

arXiv2023

HIORE: Leveraging High-order Interactions for Unified Entity Relation Extraction

Entity relation extraction consists of two sub-tasks: entity recognition and relation extraction. Existing methods either tackle these two tasks separately or unify them with word-by-word interactions. In this paper, we propose HIORE, a new method for unified entity relation extraction. The key insight is to leverage the high-order interactions, i.e., the complex association among word pairs, which contains richer information than the first-order word-by-word interactions. For this purpose, we first devise a W-shape DNN (WNet) to capture coarse-level high-order connections. Then, we build a heuristic high-order graph and further calibrate the representations with a graph neural network (GNN). Experiments on three benchmarks (ACE04, ACE05, SciERC) show that HIORE achieves the state-of-the-art performance on relation extraction and an improvement of 1.1~1.8 F1 points over the prior best unified model.

cs.CL↗

arXiv2023

How to Build an Optical Filter with an Atomic Vapor Cell

The nature of atomic vapors, their natural alignment with interatomic transitions, and their ease of use make them highly suited for spectrally narrow-banded optical filters. Atomic filters come in two flavors: a filter based on the absorption of light by the Doppler broadened atomic vapor, i.e., a notch filter, and a bandpass filter based on the transmission of resonant light caused by the Faraday effect. The notch filter uses the absorption of resonant photons to filter out a small spectral band around the atomic transition. The off-resonant part of the spectrum is fully transmitted. Atomic vapors based on the Faraday effect allow for suppression of the detuned spectral fraction. Transmission of light originates from the magnetically induced rotation of linear polarized light close to an atomic resonance. This filter constellation allows selective acceptance of specific light frequencies. In this manuscript, we discuss these two types of filters and elucidate the specialties of atomic line filters. We also present a practical guide on building such filter setups from scratch and discuss an approach to achieve an almost perfect atomic spectrum backed by theoretical calculations.

physics.atom-ph↗

arXiv2023

[OIII] 5007A Emission Line Width as a Surrogate for stellar dispersion in Type 1 AGNs?

We present a study of the relation between the [OIII] 5007A emission line width (sigma_{[OIII]}) and stellar velocity dispersion (sigma_{*}), utilizing a sample of 740 type 1 active galactic nuclei (AGNs) with high-quality spectra at redshift z < 1.0. We find the broad correlation between the core component of [OIII] emission line width (sigma_{[OIII,core]}) and sigma_{*} with a scatter of 0.11~dex for the low redshift (z < 0.1) sample; for redshift (0.3 < z < 1.0) AGNs, the scatter is larger, being 0.16~dex. We also find that the Eddington ratio (L_{bol}/L_{Edd}) may play an important role in the discrepancies between sigma_{[OIII,core]} and sigma_{*}. As the L_{bol}/L_{Edd} increases, sigma_{[OIII,core]} tends to be larger than sigma_{*}. By classifying our local sample with different minor-to-major axis ratios, we find that sigma_{*} is larger than sigma_{[OIII,core]} for those edge-on spiral galaxies. In addition, we also find that the effects of outflow strength properties such as maximum outflow velocity (V_{max}) and the broader component of [OIII] emission line width and line shift (sigma_{[OIII,out]} and V_{[OIII,out]}) may play a major role in the discrepancies between sigma_{[OIII,core]} and sigma_{*}. The discrepancies between sigma_{[OIII,core]} and sigma_{*} are larger when V_{max}, V_{[OIII,out]}, and sigma_{[OIII,out]} increase. Our results show that the outflow strengths may have significant effects on the differences between narrow-line region gas and stellar kinematics in AGNs. We suggest that caution should be taken when using sigma_{[OIII,core]} as a surrogate for sigma_{*}. In addition, the substitute of sigma_{[OIII,core]} for sigma_{*} could be used only for low luminosity AGNs.

astro-ph.GA↗

arXiv2022

NRTR: Neuron Reconstruction with Transformer from 3D Optical Microscopy Images

The neuron reconstruction from raw Optical Microscopy (OM) image stacks is the basis of neuroscience. Manual annotation and semi-automatic neuron tracing algorithms are time-consuming and inefficient. Existing deep learning neuron reconstruction methods, although demonstrating exemplary performance, greatly demand complex rule-based components. Therefore, a crucial challenge is designing an end-to-end neuron reconstruction method that makes the overall framework simpler and model training easier. We propose a Neuron Reconstruction Transformer (NRTR) that, discarding the complex rule-based components, views neuron reconstruction as a direct set-prediction problem. To the best of our knowledge, NRTR is the first image-to-set deep learning model for end-to-end neuron reconstruction. In experiments using the BigNeuron and VISoR-40 datasets, NRTR achieves excellent neuron reconstruction results for comprehensive benchmarks and outperforms competitive baselines. Results of extensive experiments indicate that NRTR is effective at showing that neuron reconstruction is viewed as a set-prediction problem, which makes end-to-end model training available.

cs.CV↗

arXiv2022

Constraining the Stochastic Gravitational Wave Background with Photometric Surveys

The detection of the Stochastic Gravitational Wave Background (SGWB) is essential for understanding black hole populations, especially for supermassive black hole binaries. The recent promising results from various Pulsar Timing Array (PTA) collaborations allude to an imminent detection. In this paper, we investigate the relative astrometric gravitational wave detection method, which can contribute to SGWB studies in the microhertz range. We consider the Roman Space Telescope and Gaia as candidates and quantitatively discuss the survey sensitivity in both the frequency and spatial domains. We emphasize the importance of survey specific constraints on performance estimates by considering mean field of view (FoV) signal subtraction and angular power spectrum binning. We conclude that if the SGWB is at a similar level as in PTA estimates, both Roman and Gaia have the potential to detect this frequency-domain power excess. However, both Roman and Gaia are subject to FoV limitations, and are unlikely to be sensitive to the spatial pattern of the SGWB.

gr-qc↗

arXiv2022

Few Clean Instances Help Denoising Distant Supervision

Existing distantly supervised relation extractors usually rely on noisy data for both model training and evaluation, which may lead to garbage-in-garbage-out systems. To alleviate the problem, we study whether a small clean dataset could help improve the quality of distantly supervised models. We show that besides getting a more convincing evaluation of models, a small clean dataset also helps us to build more robust denoising models. Specifically, we propose a new criterion for clean instance selection based on influence functions. It collects sample-level evidence for recognizing good instances (which is more informative than loss-level evidence). We also propose a teacher-student mechanism for controlling purity of intermediate results when bootstrapping the clean set. The whole approach is model-agnostic and demonstrates strong performances on both denoising real (NYT) and synthetic noisy datasets.

cs.CL↗

arXiv2022

Uncertainty and bias of cosmology and astrophysical population model from statistical dark sirens

Gravitational-wave (GW) radiation from a coalescing compact binary is a standard siren as the luminosity distance of each event can be directly measured from the amplitude of the signal. One possibility to constrain cosmology using the GW siren is to perform statistical inference on a population of binary black hole (BBH) events. In essence, this statistical method can be viewed as follows. We can modify the shape of the distribution of observed BBH events by changing cosmological parameters until it eventually matches the distribution constructed from an astrophysical population model, thereby allowing us to determine the cosmological parameters. In this work, we derive the Cramér-Rao bound for both cosmological parameters and those governing the astrophysical population model from this statistical dark siren method by examining the Fisher information contained in the event distribution. Our study provides analytical insights and enables fast yet accurate estimations of the statistical accuracy of dark siren cosmology. Furthermore, we consider the bias in cosmology due to unmodeled substructures in the merger rate and the mass distribution. We find a $1\%$ deviation in the astrophysical model can lead to a more than $1\%$ error in the Hubble constant. This could limit the accuracy of dark siren cosmology when there are more than $10^4$ BBH events detected.

astro-ph.CO↗

arXiv2022

Asynchronous Hierarchical Federated Learning

Federated Learning is a rapidly growing area of research and with various benefits and industry applications. Typical federated patterns have some intrinsic issues such as heavy server traffic, long periods of convergence, and unreliable accuracy. In this paper, we address these issues by proposing asynchronous hierarchical federated learning, in which the central server uses either the network topology or some clustering algorithm to assign clusters for workers (i.e., client devices). In each cluster, a special aggregator device is selected to enable hierarchical learning, leads to efficient communication between server and workers, so that the burden of the server can be significantly reduced. In addition, asynchronous federated learning schema is used to tolerate heterogeneity of the system and achieve fast convergence, i.e., the server aggregates the gradients from the workers weighted by a staleness parameter to update the global model, and regularized stochastic gradient descent is performed in workers, so that the instability of asynchronous learning can be alleviated. We evaluate the proposed algorithm on CIFAR-10 image classification task, the experimental results demonstrate the effectiveness of asynchronous hierarchical federated learning.

cs.LG↗

arXiv2022

Transient obscuration event captured in NGC 3227 III. Photoionization modeling of the X-ray obscuration event in 2019

A growing number of transient X-ray obscuration events in type I AGN suggest that our line-of-sight to the central engine is not always free. Multiple X-ray obscuration events have been reported in the nearby Seyfert 1.5 galaxy NGC 3227 from 2000 to 2016. In late 2019, another X-ray obscuration event was identified with Swift. Two coordinated target-of-opportunity observations with XMM-Newton, NuSTAR, and HST/COS were triggered in Nov. and Dec. 2019 to study this obscuration event. For each observation, we analyze the time-averaged X-ray spectra. We perform photoionization modeling with the SPEX code, which allows us to constrain the intrinsic continuum simultaneously with various photoionized absorption and emission components. Similar to previous transient X-ray obscuration events in NGC 3227, the one caught in late 2019 is short-lived (less than five months). If the obscurer has only one photoionized component, the two X-ray observations in late 2019 cannot be explained by the same obscurer that responds to the varying ionizing continuum. Due to the unknown geometry of the obscurer, its number density and distance to the black hole cannot be well constrained. The inferred distance covers at least two orders of magnitude, from the BLR to the dusty torus. Unlike some other X-ray obscuration events in Seyfert galaxies like NGC 5548 and NGC 3783, no prominent blueshifted broad absorption troughs were found in the 2019 HST/COS spectra of NGC 3227 when compared with archival UV spectra. This might be explained if the X-ray obscurer does not intercept our line of sight to (a significant portion of) the UV emitting region. It is not straightforward to understand the variety of the observational differences in the X-ray obscuration events observed so far. Future observations with high-quality data are needed to unveil the nature of the X-ray obscuration events. [shortend for arXiv]

astro-ph.HE↗

arXiv2022

Density profile of ambient circumnuclear medium in Seyfert 1 galaxies

The shape of the ambient circumnuclear medium (ACM) density profile can probe the history of accretion onto the central supermassive black hole in galaxies and the circumnuclear environment. However, due to the limitation of the instrument resolution, the density profiles of the ACM for most of galaxies remain largely unknown. In this work, we propose a novel method to measure the ACM density profile of active galactic nucleus (AGN) by the equilibrium between the radiation pressure on the warm absorbers (WAs, a type of AGN outflows) and the drag pressure from the ACM. We study the correlation between the outflow velocity and ionization parameter of WAs in each of the five Seyfert 1 galaxies (NGC 3227, NGC 3783, NGC 4051, NGC 4593, and NGC 5548), inferring that the density profile of the ACM is between n\propto r^-1.7 and n \propto r^-2.15 (n is number density and r is distance) from 0.01 pc to pc scales in these five AGNs. Our results indicate that the ACM density profile in Seyfert 1 galaxies is steeper than the prediction by the spherically symmetric Bondi accretion model and the simulated results of the hot accretion flow, but more in line with the prediction by the standard thin disk model.

astro-ph.GA↗

arXiv2021

YACLC: A Chinese Learner Corpus with Multidimensional Annotation

Learner corpus collects language data produced by L2 learners, that is second or foreign-language learners. This resource is of great relevance for second language acquisition research, foreign-language teaching, and automatic grammatical error correction. However, there is little focus on learner corpus for Chinese as Foreign Language (CFL) learners. Therefore, we propose to construct a large-scale, multidimensional annotated Chinese learner corpus. To construct the corpus, we first obtain a large number of topic-rich texts generated by CFL learners. Then we design an annotation scheme including a sentence acceptability score as well as grammatical error and fluency-based corrections. We build a crowdsourcing platform to perform the annotation effectively (https://yaclc.wenmind.net). We name the corpus YACLC (Yet Another Chinese Learner Corpus) and release it as part of the CUGE benchmark (http://cuge.baai.ac.cn). By analyzing the original sentences and annotations in the corpus, we found that YACLC has a considerable size and very high annotation quality. We hope this corpus can further enhance the studies on Chinese International Education and Chinese automatic grammatical error correction.

cs.CL↗

arXiv2021

Transient obscuration event captured in NGC~3227 II. Warm absorbers and obscuration events in archival XMM-Newton and NuSTAR observations

The relation between warm absorber (WA) outflows of AGN and nuclear obscuration activities caused by optically-thick clouds (obscurers) crossing the line of sight is unclear. NGC 3227 is a suitable target to study the properties of both WAs and obscurers, because it matches the following selection criteria: WAs in both ultraviolet (UV) and X-rays, suitably variable, bright in UV and X-rays, good archival spectra for comparing with the obscured spectra. To investigate WAs and obscurers of NGC~3227, we used a broadband spectral-energy-distribution model built in our Paper I and the photoionization code of SPEX software to fit archival XMM-Newton and NuSTAR observations in 2006 and 2016. Using unobscured observations, we find four WAs with different ionization states (log$ξ$ [erg cm/s]~-1.0, 2.0, 2.5, 3.0). The highest-ionization WA has a higher hydrogen column density (~$10^{22}$/cm$^2$) than the other three WAs (~$10^{21}$/cm$^2$). Their outflow velocities range from 100 to 1300 km/s, and show a positive correlation with the ionization parameter. These WAs are estimated to be between the outer broad-line-region (BLR) and the narrow line region. Besides, we find an X-ray obscuration event in 2006, which was missed by previous studies. It can be explained by a single obscurer. We also study the previously published obscuration event in 2016, which needs two obscurers in the fit. A high-ionization obscurer (log$ξ$~2.80; covering factor $C_f$~30%) only appears in 2016, which has a high column density (~$10^{23}$/cm$^2$). A low-ionization obscurer (log$ξ$~1.0-1.9; $C_f$~20%-50%) exists in both 2006 and 2016, which has a lower column density (~$10^{22}$/cm$^2$). These obscurers are estimated to be in the BLR by their crossing time of transverse motions. The obscurers and WAs of NGC 3227 have different distances and number densities, which indicate that they might have different origins.

astro-ph.GA↗

arXiv2021

UniRE: A Unified Label Space for Entity Relation Extraction

Many joint entity relation extraction models setup two separated label spaces for the two sub-tasks (i.e., entity detection and relation classification). We argue that this setting may hinder the information interaction between entities and relations. In this work, we propose to eliminate the different treatment on the two sub-tasks' label spaces. The input of our model is a table containing all word pairs from a sentence. Entities and relations are represented by squares and rectangles in the table. We apply a unified classifier to predict each cell's label, which unifies the learning of two sub-tasks. For testing, an effective (yet fast) approximate decoder is proposed for finding squares and rectangles from tables. Experiments on three benchmarks (ACE04, ACE05, SciERC) show that, using only half the number of parameters, our model achieves competitive accuracy with the best extractor, and is faster.

cs.CL↗

arXiv2021

Identifying Type II Strongly Lensed Gravitational-Wave Images in Third-Generation Gravitational-Wave Detectors

Strong gravitational lensing is a gravitational wave (GW) propagation effect that influences the inferred GW source parameters and the cosmological environment. Identifying strongly-lensed GW images is challenging as waveform amplitude magnification is degenerate with a shift in the source intrinsic mass and redshift. However, even in the geometric-optics limit, type II strongly-lensed images cannot be fully matched by type I (or unlensed) waveform templates, especially with large binary mass ratios and orbital inclination angles. We propose to use this mismatch to distinguish individual type II images. Using planned noise spectra of Cosmic Explorer, Einstein Telescope and LIGO Voyager, we show that a significant fraction of type II images can be distinguished from unlensed sources, given sufficient SNR ($\sim 30$). Incorporating models on GW source population and lens population, we predict that the yearly detection rate of lensed GW sources with detectable type II images is 172.2, 118.2 and 27.4 for CE, ET and LIGO Voyager, respectively. Among these detectable events, 33.1%, 7.3% and 0.22% will be distinguishable via their type II images with a log Bayes factor larger than 10. We conclude that such distinguishable events are likely to appear in the third-generation detector catalog; our strategy will significantly supplement existing strong lensing search strategies.

gr-qc↗

arXiv2021

Gravitational Wave Detection with Photometric Surveys

Gravitational wave (GW) detections have enriched our understanding of the universe. To date, all single-source GW events were found by interferometer-type detectors. We study a detection method using astrometric solutions from photometric surveys and demonstrate that it offers a highly flexible frequency range, uniquely complementing existing detection methods. From repeated point-source astrometric measurements, we may extract GW-induced deflections and infer wave parameters. This method can be applied to any photometric surveys measuring relative astrometry. We show that high-cadence observations of the galactic bulge, such as offered by the Roman Space Telescope's Exoplanet MicroLensing (EML) survey, can be a potent GW probe with complementary frequency range to Gaia, pulsar timing arrays (PTAs), and the Laser Interferometer Space Antenna (LISA). We calculate that the Roman EML survey is sensitive to GWs with frequencies ranging from $7.7\times10^{-8}$Hz to $5.6\times10^{-4}$Hz, which opens up a unique GW observing window for supermassive black hole binaries and their waveform evolution. While the detection threshold assuming the currently expected performance proves too high for detecting individual GWs in light of the expected supermassive black hole binary population distribution, we show that binaries with chirp mass $M_c>10^{8.3}~M_\odot$ out to 100 Mpc can be detected if the telescope is able to achieve an astrometric accuracy of 0.11 mas. To confidently detect binaries with $M_c>10^{7}~M_\odot$ out to 50 Mpc, a factor of 100 sensitivity improvement is required. We propose several improvement strategies, including recovering the mean astrometric deflection and increasing astrometric accuracy, number of observed stars, field-of-view size, and observational cadence. We discuss how other existing and planned photometric surveys could contribute to detecting GWs via astrometry.

gr-qc↗

arXiv2020

Stock2Vec: A Hybrid Deep Learning Framework for Stock Market Prediction with Representation Learning and Temporal Convolutional Network

We have proposed to develop a global hybrid deep learning framework to predict the daily prices in the stock market. With representation learning, we derived an embedding called Stock2Vec, which gives us insight for the relationship among different stocks, while the temporal convolutional layers are used for automatically capturing effective temporal patterns both within and across series. Evaluated on S&P 500, our hybrid framework integrates both advantages and achieves better performance on the stock price prediction task than several popular benchmarked models.

q-fin.ST↗

arXiv2020

tACS Facilitates Flickering Driving by Boosting Steady-State Visual Evoked Potentials

There has become of increasing interest in transcranial alternating current stimulation (tACS) since its inception nearly a decade ago. tACS in modulating brain state is an active area of research and has been demonstrated effective in various neuropsychological and clinical domains. In the visual domain, much effort has been dedicated to brain rhythms and rhythmic stimulation, i.e., tACS. However, little is known about the interplay between the rhythmic stimulation and visual stimulation. Here, we used steady-state visual evoked potential (SSVEP), induced by flickering driving as a widely used technique for frequency-tagging, to investigate the aftereffect of tACS in healthy human subjects. Seven blocks of 64-channel electroencephalogram were recorded before and after the administration of 20-min 10-Hz tACS, while subjects performed several blocks of SSVEP tasks. We characterized the physiological properties of tACS aftereffect by comparing and validating the temporal, spatial, spatiotemporal and signal-to-noise ratio (SNR) patterns between and within blocks in real tACS and sham tACS. Our result revealed that tACS boosted the 10-Hz SSVEP significantly. Besides, the aftereffect on SSVEP was mitigated with time and lasted up to 5 min. Our results demonstrate the feasibility of facilitating the flickering driving by external rhythmic stimulation and open a new possibility to alter the brain state in a direction by noninvasive transcranial brain stimulation.

q-bio.NC↗