arXiv subjects

Jiaxuan Li

Publications and source records attributed to Jiaxuan Li.

89 records · Page 5Linked to original sources

arXiv2022

A unified information-theoretic model of EEG signatures of human language processing

We advance an information-theoretic model of human language processing in the brain, in which incoming linguistic input is processed at two levels, in terms of a heuristic interpretation and in terms of error correction. We propose that these two kinds of information processing have distinct electroencephalographic signatures, corresponding to the well-documented N400 and P600 components of language-related event-related potentials (ERPs). Formally, we show that the information content (surprisal) of a word in context can be decomposed into two quantities: (A) heuristic surprise, which signals processing difficulty of word given its inferred context, and corresponds with the N400 signal; and (B) discrepancy signal, which reflects divergence between the true context and the inferred context, and corresponds to the P600 signal. Both of these quantities can be estimated using modern NLP techniques. We validate our theory by successfully simulating ERP patterns elicited by a variety of linguistic manipulations in previously-reported experimental data from Ryskin et al. (2021). Our theory is in principle compatible with traditional cognitive theories assuming a `good-enough' heuristic interpretation stage, but with precise information-theoretic formulation.

cs.CL↗

arXiv2022

Counterfactual reasoning: Do language models need world knowledge for causal understanding?

Current pre-trained language models have enabled remarkable improvements in downstream tasks, but it remains difficult to distinguish effects of statistical correlation from more systematic logical reasoning grounded on understanding of the real world. In this paper we tease these factors apart by leveraging counterfactual conditionals, which force language models to predict unusual consequences based on hypothetical propositions. We introduce a set of tests drawn from psycholinguistic experiments, as well as larger-scale controlled datasets, to probe counterfactual predictions from a variety of popular pre-trained language models. We find that models are consistently able to override real-world knowledge in counterfactual scenarios, and that this effect is more robust in case of stronger baseline world knowledge -- however, we also find that for most models this effect appears largely to be driven by simple lexical cues. When we mitigate effects of both world knowledge and lexical cues to test knowledge of linguistic nuances of counterfactuals, we find that only GPT-3 shows sensitivity to these nuances, though this sensitivity is also non-trivially impacted by lexical associative factors.

cs.CL↗

arXiv2022

Tidal Distortions in NGC1052-DF2 and NGC1052-DF4: Independent Evidence for a Lack of Dark Matter

Two ultra diffuse galaxies in the same group, NGC1052-DF2 and NGC1052-DF4, have been found to have little or no dark matter and to host unusually luminous globular cluster populations. Such low mass diffuse objects in a group environment are easily disrupted and are expected to show evidence of tidal distortions. In this work we present deep new imaging of the NGC1052 group, obtained with the Dragonfly Telephoto Array, to test this hypothesis. We find that both galaxies show strong position angle twists and are significantly more elongated at their outskirts than in their interiors. The group's central massive elliptical NGC1052 is the most likely source of these tidal disturbances. The observed distortions imply that the galaxies have a low total mass or are very close to NGC1052. Considering constraints on the galaxies' relative distances, we infer that the dark matter halo masses of these galaxies cannot be much greater than their stellar masses. Calculating pericenters from the distortions, we find that the galaxies are on highly elliptical orbits, with a ratio of pericenter to present-day radius Rperi/R0~0.1 if the galaxies are dark matter-free and Rperi/R0~0.01 if they have a normal dark halo. Our findings provide strong evidence, independent of kinematic constraints, that both galaxies are dark matter deficient. Furthermore, the similarity of the tidal features in NGC1052-DF2 and NGC1052-DF4 strongly suggests that they arose at comparable distances from NGC1052. In Appendix A, we describe sbcontrast, a robust method to determine the surface brightness limit of images.

astro-ph.GA↗

arXiv2022

The Nature of Low Surface Brightness Galaxies in the Hyper Suprime-Cam Survey

We present the statistical redshift distribution of a large sample of low surface brightness (LSB) galaxies identified in the first 200 deg$^2$ of the Hyper Suprime-Cam Strategic Survey Program. Through cross-correlation with the NASA-SDSS Atlas, we find that the majority of objects lie within z<0.15 or ~ 500 Mpc, yielding a mass range of $M_*$ ~ $10^7-10^9$ M_sun and size range of $r_{\rm{eff,g}}$ ~ 1-8 kpc. We find a peak in the distance distribution within 100 Mpc, corresponding mostly to ~ $10^7$ M_sun galaxies that fall on the known mass-size relation. There is also a tail in the redshift distribution out to z~0.15, comprising more massive ($M_*=10^8-10^9$ M_sun) galaxies at the larger end of our size range. We see tentative evidence that at the higher-mass end ($M_* > 10^8$ M_sun) the LSB galaxies do not form a smooth extension of the mass-size relation of higher surface brightness galaxies, perhaps suggesting that the LSB galaxy population is distinct in its formation path.

astro-ph.GA↗

arXiv2022

Anti-Asian Hate Speech Detection via Data Augmented Semantic Relation Inference

With the spreading of hate speech on social media in recent years, automatic detection of hate speech is becoming a crucial task and has attracted attention from various communities. This task aims to recognize online posts (e.g., tweets) that contain hateful information. The peculiarities of languages in social media, such as short and poorly written content, lead to the difficulty of learning semantics and capturing discriminative features of hate speech. Previous studies have utilized additional useful resources, such as sentiment hashtags, to improve the performance of hate speech detection. Hashtags are added as input features serving either as sentiment-lexicons or extra context information. However, our close investigation shows that directly leveraging these features without considering their context may introduce noise to classifiers. In this paper, we propose a novel approach to leverage sentiment hashtags to enhance hate speech detection in a natural language inference framework. We design a novel framework SRIC that simultaneously performs two tasks: (1) semantic relation inference between online posts and sentiment hashtags, and (2) sentiment classification on these posts. The semantic relation inference aims to encourage the model to encode sentiment-indicative information into representations of online posts. We conduct extensive experiments on two real-world datasets and demonstrate the effectiveness of our proposed framework compared with state-of-the-art representation learning models.

cs.CL↗

arXiv2022

Multi-scale Sparse Representation-Based Shadow Inpainting for Retinal OCT Images

Inpainting shadowed regions cast by superficial blood vessels in retinal optical coherence tomography (OCT) images is critical for accurate and robust machine analysis and clinical diagnosis. Traditional sequence-based approaches such as propagating neighboring information to gradually fill in the missing regions are cost-effective. But they generate less satisfactory outcomes when dealing with larger missing regions and texture-rich structures. Emerging deep learning-based methods such as encoder-decoder networks have shown promising results in natural image inpainting tasks. However, they typically need a long computational time for network training in addition to the high demand on the size of datasets, which makes it difficult to be applied on often small medical datasets. To address these challenges, we propose a novel multi-scale shadow inpainting framework for OCT images by synergically applying sparse representation and deep learning: sparse representation is used to extract features from a small amount of training images for further inpainting and to regularize the image after the multi-scale image fusion, while convolutional neural network (CNN) is employed to enhance the image quality. During the image inpainting, we divide preprocessed input images into different branches based on the shadow width to harvest complementary information from different scales. Finally, a sparse representation-based regularizing module is designed to refine the generated contents after multi-scale feature aggregation. Experiments are conducted to compare our proposal versus both traditional and deep learning-based techniques on synthetic and real-world shadows. Results demonstrate that our proposed method achieves favorable image inpainting in terms of visual quality and quantitative metrics, especially when wide shadows are presented.

cs.CV↗

arXiv2022

Cold Gas in Massive Galaxies as A Critical Test of Black Hole Feedback Models

Black hole feedback has been widely implemented as the key recipe to quench star formation in massive galaxies in modern semi-analytic models and hydrodynamical simulations. As the theoretical details surrounding the accretion and feedback of black holes continue to be refined, various feedback models have been implemented across simulations, with notable differences in their outcomes. Yet, most of these simulations have successfully reproduced some observations, such as stellar mass function and star formation rate density in the local Universe. We use the recent observation on the change of neutral hydrogen gas mass (including both ${\rm H_2}$ and ${\rm HI}$) with star formation rate of massive central disc galaxies as a critical constraint of black hole feedback models across several simulations. We find that the predictions of IllustrisTNG agree with the observations much better than the other models tested in this work. This favors IllustrisTNG's treatment of active galactic nuclei - where kinetic winds are driven by black holes at low accretion rates - as more plausible amongst those we test. In turn, this also indirectly supports the idea that the massive central disc galaxy population in the local Universe was likely quenched by AGN feedback.

astro-ph.GA↗

arXiv2021

NGC5846-UDG1: A galaxy formed mostly by star formation in massive, extremely dense clumps of gas

It has been shown that ultra-diffuse galaxies (UDGs) have higher specific frequencies of globular clusters on average than other dwarf galaxies with similar luminosities. The UDG NGC5846-UDG1 is among the most extreme examples of globular cluster-rich galaxies found so far. Here we present new Hubble Space Telescope (HST) observations and analysis of this galaxy and its globular cluster system. We find that NGC5846-UDG1 hosts $54 \pm 9$ globular clusters, three to four times more than any previously known galaxy with a similar luminosity, and higher than reported in previous studies. With a galaxy luminosity of $L_{V,\mathrm{gal}} \approx 6 \times 10^7\,{\rm L}_{\odot}$ ($M_\star \approx 1.2 \times 10^8\,{\rm M}_\odot$) and a total globular cluster luminosity of $L_{V,\mathrm{GCs}} \approx 7.6 \times 10^6\,{\rm L}_{\odot}$, we find that the clusters currently comprise $\sim 13 \%$ of the total light. Taking into account the effects of mass loss from clusters during their formation and throughout their lifetime, we infer that most of the stars in the galaxy likely formed in globular clusters, and very little to no "normal" low-density star formation occurred. This result implies that the most extreme conditions during early galaxy formation promoted star formation in massive and dense clumps, in contrast to the dispersed star formation observed in galaxies today.

astro-ph.GA↗

arXiv2021

A Method To Characterize the Wide-Angle Point Spread Function of Astronomical Images

Uncertainty in the wide-angle Point Spread Function (PSF) at large angles (tens of arcseconds and beyond) is one of the dominant sources of error in a number of important quantities in observational astronomy. Examples include the stellar mass and shape of galactic halos and the maximum extent of starlight in the disks of nearby galaxies. However, modeling the wide-angle PSF has long been a challenge in astronomical imaging. In this paper, we present a self-consistent method to model the wide-angle PSF in images. Scattered light from multiple bright stars is fitted simultaneously with a background model to characterize the extended wing of the PSF using a Bayesian framework operating on pixel-by-pixel level. The method is demonstrated using our software elderflower and is applied to data from the Dragonfly Telephoto Array to model its PSF out to 20-25 arcminutes. We compare the wide-angle PSF of Dragonfly to that of a number of other telescopes, including the SDSS PSF, and show that on scales of arcminutes the scattered light in the Dragonfly PSF is markedly lower than that of other wide-field imaging telescopes. The energy in the wings of the Dragonfly point-spread function is sufficiently low that optical cleanliness plays an important role in defining the PSF. This component of the PSF can be modelled accurately, highlighting the power of our self-contained approach.

astro-ph.IM↗

arXiv2021

Reaching for the Edge I: Probing the Outskirts of Massive Galaxies with HSC, DECaLS, SDSS, and Dragonfly

The outer light (stellar halos) of massive galaxies has recently emerged as a possible low scatter tracer of dark matter halo mass. To test the robustness of outer light measurements across different data sets, we compare the surface brightness profiles of massive galaxies using four independent data sets: the Hyper Suprime-Cam survey (HSC), the Dark Energy Camera Legacy Survey (DECaLS), the Sloan Digital Sky Survey (SDSS), and the Dragonfly Wide Field Survey (Dragonfly). We use customized pipelines for HSC and DECaLS to achieve better sky background subtraction. For galaxies at $z<0.05$, Dragonfly has the best control of systematics, reaching surface brightness levels of $μ_r \sim 30$ mag/arcsec$^{2}$. At $0.19 200$ kpc without significant bias. At $0.19<z<0.50$, DECaLS and HSC measurements of the stellar mass contained within 100 kpc agree within 0.05 dex. Finally, we use weak gravitational lensing to show that measurements of outer light with DECaLS at $0.19<z<0.50$ show a similar promise as HSC as a low scatter proxy of halo mass. The tests and results from this paper represent an important step forward for accurate measurements of the outer light of massive galaxies and demonstrate that outer light measurements from DECam imaging will be a promising method for finding galaxy clusters for DES and DESI.

astro-ph.GA↗

arXiv2021

Multi-scale GCN-assisted two-stage network for joint segmentation of retinal layers and disc in peripapillary OCT images

An accurate and automated tissue segmentation algorithm for retinal optical coherence tomography (OCT) images is crucial for the diagnosis of glaucoma. However, due to the presence of the optic disc, the anatomical structure of the peripapillary region of the retina is complicated and is challenging for segmentation. To address this issue, we developed a novel graph convolutional network (GCN)-assisted two-stage framework to simultaneously label the nine retinal layers and the optic disc. Specifically, a multi-scale global reasoning module is inserted between the encoder and decoder of a U-shape neural network to exploit anatomical prior knowledge and perform spatial reasoning. We conducted experiments on human peripapillary retinal OCT images. The Dice score of the proposed segmentation network is 0.820$\pm$0.001 and the pixel accuracy is 0.830$\pm$0.002, both of which outperform those from other state-of-the-art techniques.

eess.IV↗

arXiv2021

The Dragonfly Wide Field Survey. II. Accurate Total Luminosities and Colors of Nearby Massive Galaxies and Implications for the Galaxy Stellar Mass Function

Stellar mass estimates of massive galaxies are susceptible to systematic errors in their photometry, due to their extended light profiles. In this study we use data from the Dragonfly Wide Field Survey (DWFS) to accurately measure the total luminosities and colors of nearby massive galaxies. The low surface brightness limits of the survey ($μ_g \approx $ 31 mag arcsec $^{-2}$ on a one arcmin scale) allows us to implement a method, based on integrating the 1-D surface brightness profile, that is minimally dependent on any parameterization. We construct a sample of 1188 massive galaxies with $\log M_*/M_\odot > 10.75$ based on the Galaxy Mass and Assembly (GAMA) survey and measure their total luminosities and $g-r$ colors. We then compare our measurements to various established methods applied to imaging from the Sloan Digital Sky Survey (SDSS), focusing on those favored by the GAMA survey. In general, we find that galaxies are brighter in the $r$ band by an average of ${\sim}0.05$ mag and bluer in $g-r$ colors by $\sim 0.06$ mag compared to the GAMA measurements. These two differences have opposite effects on the stellar mass estimates. The total luminosities are larger by $5\%$ but the mass-to-light ratios are lower by $\sim 10\%$. The combined effect is that the stellar mass estimate of massive galaxies decreases by $7\%$. This, in turn, implies a small change in number density of massive galaxies: $\leq 30\%$ at $\log M_*/M_\odot \geq 11$.

astro-ph.GA↗

arXiv2020

A Numerical Study on the Effects of Heterogeneity, Anisotropy, and Station Coverage on the Compensated Linear Vector Dipole Component of Deep Earthquake Moment Tensors

The moment tensors of a large portion of deep earthquakes show apparent non-double-couple (non-DC) components. Previously, the observed apparent non-DC values in deep earthquakes have been attributed to different mechanisms such as complex source processes or complicated source medium structures. In this paper, we focused on evaluating the second mechanism. We investigated the effect of slab heterogeneity, supra-slab anisotropic structure, intra-slab weakly anisotropic structure (e.g., the purported existence of the metastable olivine wedge), and non-uniform station coverage, on the non-DC radiation patterns of deep earthquakes using our 3-dimensional elastic finite-difference modeling and full-waveform inversion of moment tensors. We found that these investigated issues cannot cause the observed non-double-couple radiation patterns and the in-situ structure with strong S-wave anisotropy near to the earthquake focus is the simplest way to account for the apparent non-DC components in the radiation patterns of deep earthquakes.

physics.geo-ph↗

arXiv2020

Multi-resolution filtering: an empirical method for isolating faint, extended emission in Dragonfly data and other low resolution images

We describe an empirical, self-contained method to isolate faint, large-scale emission in imaging data of low spatial resolution. Multi-resolution filtering (MRF) uses independent data of superior spatial resolution and point source depth to create a model for all compact and high surface brightness objects in the field. This model is convolved with an appropriate kernel and subtracted from the low resolution image. The halos of bright stars are removed in a separate step and artifacts are masked. The resulting image only contains extended emission fainter than a pre-defined surface brightness limit. The method was developed for the Dragonfly Telephoto Array, which produces images that have excellent low surface brightness sensitivity but poor spatial resolution. We demonstrate the MRF technique using Dragonfly images of a satellite of the spiral galaxy M101, the tidal debris surrounding M51, two ultra-diffuse galaxies in the Coma cluster, and the galaxy NGC5907. As part of the analysis we present a newly-identified very faint galaxy in the filtered Dragonfly image of the M101 field. We also discuss variations of the technique for cases when no low resolution data are available (self-MRF and cross-MRF). The method is implemented in mrf, an open-source MIT licensed Python package.

astro-ph.GA↗

arXiv2020

The Dragonfly Wide Field Survey. I. Telescope, Survey Design and Data Characterization

We present a description of the Dragonfly Wide Field Survey (DWFS), a deep photometric survey of a wide area of sky. The DWFS covers 330 $\mathrm{deg}^2$ in the equatorial GAMA fields and the Stripe 82 fields in the SDSS $g$ and $r$ bands. It is carried out with the 48-lens Dragonfly Telephoto Array, a telescope that is optimized for the detection of low surface brightness emission. The main goal of the survey is to study the dwarf galaxy population beyond the Local Group. In this paper, we describe the survey design and show early results. We reach $1σ$ depths of $μ_g\approx 31$ mag arcsec$^{-2}$ on arcminute scales and show that Milky Way satellites such as Sextans, Bootes, and Ursa Major should be detectable out to $D\gtrsim 10$ Mpc. We also provide an overview of the elements and operation of the 48-lens Dragonfly telescope and a detailed description of its data reduction pipeline. The pipeline is fully automated, with individual frames subjected to a rigorous series of quality tests. The sky subtraction is performed in two stages, ensuring that emission features with spatial scales up to $\sim 0.^{\circ}9 \times 0.^{\circ}6$ are preserved. The DWFS provides unparalleled sensitivity to low surface brightness features on arcminute scales.

astro-ph.GA↗

arXiv2018

Generation of a Stochastic Binary Field That Fits a Given Heterogeneity Power Spectrum

Incomplete binary mixing of two components can form a heterogeneous assemblage in space. The heterogeneity power spectrum of the assemblage can be frequently obtained. However, it is unknown if a stochastic binary field exists to generate the observed spectrum. We propose a novel and powerful constructive procedure for this purpose. The procedure allows us not only to test whether certain binary mixing is feasible but also to tightly constrain the properties of the mixing components and the modal proportion. The method should find wide applications in many branches of geosciences.

physics.geo-ph↗

arXiv2018

Moment Inequalities in the Context of Simulated and Predicted Variables

This paper explores the effects of simulated moments on the performance of inference methods based on moment inequalities. Commonly used confidence sets for parameters are level sets of criterion functions whose boundary points may depend on sample moments in an irregular manner. Due to this feature, simulation errors can affect the performance of inference in non-standard ways. In particular, a (first-order) bias due to the simulation errors may remain in the estimated boundary of the confidence set. We demonstrate, through Monte Carlo experiments, that simulation errors can significantly reduce the coverage probabilities of confidence sets in small samples. The size distortion is particularly severe when the number of inequality restrictions is large. These results highlight the danger of ignoring the sampling variations due to the simulation errors in moment inequality models. Similar issues arise when using predicted variables in moment inequalities models. We propose a method for properly correcting for these variations based on regularizing the intersection of moments in parameter space, and we show that our proposed method performs well theoretically and in practice.

econ.EM↗