arXiv subjects

Kai Zhang

Publications and source records attributed to Kai Zhang.

At least 523 records · Page 29Linked to original sources

arXiv2022

Self-supervised Transparent Liquid Segmentation for Robotic Pouring

Liquid state estimation is important for robotics tasks such as pouring; however, estimating the state of transparent liquids is a challenging problem. We propose a novel segmentation pipeline that can segment transparent liquids such as water from a static, RGB image without requiring any manual annotations or heating of the liquid for training. Instead, we use a generative model that is capable of translating images of colored liquids into synthetically generated transparent liquid images, trained only on an unpaired dataset of colored and transparent liquid images. Segmentation labels of colored liquids are obtained automatically using background subtraction. Our experiments show that we are able to accurately predict a segmentation mask for transparent liquids without requiring any manual annotations. We demonstrate the utility of transparent liquid segmentation in a robotic pouring task that controls pouring by perceiving the liquid height in a transparent cup. Accompanying video and supplementary materials can be found

cs.RO↗

arXiv2022

Quantifying radial migration in the Milky Way: Inefficient over short timescales but essential to the very outer disc beyond ~15 kpc

Stellar radial migration plays an important role in reshaping a galaxy's structure and the radial distribution of stellar population properties. In this work, we revisit reported observational evidence for radial migration and quantify its strength using the age--[Fe/H] distribution of stars across the Milky Way with APOGEE data. We find a broken age--[Fe/H] relation in the Galactic disc at $r>6$ kpc, with a more pronounced break at larger radii. To quantify the strength of radial migration, we assume stars born at each radius have a unique age and metallicity, and then decompose the metallicity distribution function (MDF) of mono-age young populations into different Gaussian components that originated from various birth radii at $r_{\rm birth}<13$ kpc. We find that, at ages of 2 and 3 Gyr, roughly half the stars were formed within 1 kpc of their present radius, and very few stars ($<5$%) were formed more than 4 kpc away from their present radius. These results suggest limited short distance radial migration and inefficient long distance migration in the Milky Way during the last 3 Gyr. In the very outer disc beyond 15~kpc, the observed age--[Fe/H] distribution is consistent with the prediction of pure radial migration from smaller radii, suggesting a migration origin of the very outer disc. We also estimate intrinsic metallicity gradients at ages of 2 and 3 Gyr of $-0.061$ dex kpc$^{-1}$ and $-0.063$ dex kpc$^{-1}$, respectively.

astro-ph.GA↗

arXiv2022

Boundary localization of transmission eigenfunctions in spherically stratified media

Consider the transmission eigenvalue problem for $u \in H^1(Ω)$ and $v\in H^1(Ω)$ associated with $(Ω; σ, \mathbf{n}^2)$, where $Ω$ is a ball in $\mathbb{R}^N$, $N=2,3$. If $σ$ and $\mathbf{n}$ are both radially symmetric, namely they are functions of the radial parameter $r$ only, we show that there exists a sequence of transmission eigenfunctions $\{u_m, v_m\}_{m\in\mathbb{N}}$ associated with $k_m\rightarrow+\infty$ as $m\rightarrow+\infty$ such that the $L^2$-energies of $v_m$'s are concentrated around $\partialΩ$. If $σ$ and $\mathbf{n}$ are both constant, we show the existence of transmission eigenfunctions $\{u_j, v_j\}_{j\in\mathbb{N}}$ such that both $u_j$ and $v_j$ are localized around $\partialΩ$. Our results extend the recent studies in [15,16]. Through numerics, we also discuss the effects of the medium parameters, namely $σ$ and $\mathbf{n}$, on the geometric patterns of the transmission eigenfunctions.

math.AP↗

arXiv2022

An Adaptive Neuro-Fuzzy System with Integrated Feature Selection and Rule Extraction for High-Dimensional Classification Problems

A major limitation of fuzzy or neuro-fuzzy systems is their failure to deal with high-dimensional datasets. This happens primarily due to the use of T-norm, particularly, product or minimum (or a softer version of it). Thus, there are hardly any work dealing with datasets with dimensions more than hundred or so. Here, we propose a neuro-fuzzy framework that can handle datasets with dimensions even more than 7000! In this context, we propose an adaptive softmin (Ada-softmin) which effectively overcomes the drawbacks of ``numeric underflow" and ``fake minimum" that arise for existing fuzzy systems while dealing with high-dimensional problems. We call it an Adaptive Takagi-Sugeno-Kang (AdaTSK) fuzzy system. We then equip the AdaTSK system to perform feature selection and rule extraction in an integrated manner. In this context, a novel gate function is introduced and embedded only in the consequent parts, which can determine the useful features and rules, in two successive phases of learning. Unlike conventional fuzzy rule bases, we design an enhanced fuzzy rule base (En-FRB), which maintains adequate rules but does not grow the number of rules exponentially with dimension that typically happens for fuzzy neural networks. The integrated Feature Selection and Rule Extraction AdaTSK (FSRE-AdaTSK) system consists of three sequential phases: (i) feature selection, (ii) rule extraction, and (iii) fine tuning. The effectiveness of the FSRE-AdaTSK is demonstrated on 19 datasets of which five are in more than 2000 dimension including two with dimension greater than 7000. This may be the first time fuzzy systems are realized for classification involving more than 7000 input features.

cs.LG↗

arXiv2022

${W}^{2,p}$ Estimates for Elliptic Equations on $C^{1,α}$ Domains

In this paper, a new method is represented to investigate boundary $W^{2,p}$ estimates for elliptic equations, which is, roughly speaking, to derive boundary $W^{2,p}$ estimates from interior $W^{2,p}$ estimates by Whitney decomposition. Using it, $W^{2,p}$ estimates on $C^{1,α}$ domains are obtained for nondivergence form linear elliptic equations and further more, fully nonlinear elliptic equations are also considered.

math.AP↗

arXiv2022

Angular clustering properties of the DESI QSO target selection using DR9 Legacy Imaging Surveys

The quasar target selection for the upcoming survey of the Dark Energy Spectroscopic Instrument (DESI) will be fixed for the next five years. The aim of this work is to validate the quasar selection by studying the impact of imaging systematics as well as stellar and galactic contaminants, and to develop a procedure to mitigate them. Density fluctuations of quasar targets are found to be related to photometric properties such as seeing and depth of the Data Release 9 of the DESI Legacy Imaging Surveys. To model this complex relation, we explore machine learning algorithms (Random Forest and Multi-Layer Perceptron) as an alternative to the standard linear regression. Splitting the footprint of the Legacy Imaging Surveys into three regions according to photometric properties, we perform an independent analysis in each region, validating our method using eBOSS EZ-mocks. The mitigation procedure is tested by comparing the angular correlation of the corrected target selection on each photometric region to the angular correlation function obtained using quasars from the Sloan Digital Sky Survey (SDSS)Data Release 16. With our procedure, we recover a similar level of correlation between DESI quasar targets and SDSS quasars in two thirds of the total footprint and we show that the excess of correlation in the remaining area is due to a stellar contamination which should be removed with DESI spectroscopic data. We derive the Limber parameters in our three imaging regions and compare them to previous measurements from SDSS and the 2dF QSO Redshift Survey.

astro-ph.CO↗

arXiv2022

Wake interactions between two side-by-side circular cylinders with different sizes

Flows over two side-by-side circular cylinders exhibit fascinating flow physics due to complex interactions between the coupled wakes. However, their mutual interference effects have not been elucidated in a quantitative manner thus far. In this paper, we create a mismatch between the two wakes by introducing a size difference in the cylinder pair, such that the effects of one wake on the other can be distinguished. Depending on the size and gap ratios between the two cylinders, the coupled wake exhibits distinct dynamical features including the quasi-periodic flow, synchronized flow, and chaotic flow. Through advanced spectral analysis of lift coefficients and dynamic mode decomposition of the flow fields, we reveal that the quasi-periodic flows are mainly composed of two primary frequencies associated with vortex shedding in the near wakes of the two cylinders. Both wakes impose their own frequencies on the other, resulting in the beating phenomenon in the lift coefficients. The triad interactions between the two wakes generate the sideband frequencies, which are associated with modal structures that are mostly active in the far wake. The transition from quasi-periodicity to synchronization is dominated by the vortex shedding behind the larger cylinder, to which the wake of the smaller cylinder locks in. These results reveal new insights on the coupled wakes of two circular cylinders, and are pivotal for understanding more general wake interaction problems.

physics.flu-dyn↗

arXiv2021

Gendered Differences in Face Recognition Accuracy Explained by Hairstyles, Makeup, and Facial Morphology

Media reports have accused face recognition of being ''biased'', ''sexist'' and ''racist''. There is consensus in the research literature that face recognition accuracy is lower for females, who often have both a higher false match rate and a higher false non-match rate. However, there is little published research aimed at identifying the cause of lower accuracy for females. For instance, the 2019 Face Recognition Vendor Test that documents lower female accuracy across a broad range of algorithms and datasets also lists ''Analyze cause and effect'' under the heading ''What we did not do''. We present the first experimental analysis to identify major causes of lower face recognition accuracy for females on datasets where previous research has observed this result. Controlling for equal amount of visible face in the test images mitigates the apparent higher false non-match rate for females. Additional analysis shows that makeup-balanced datasets further improves females to achieve lower false non-match rates. Finally, a clustering experiment suggests that images of two different females are inherently more similar than of two different males, potentially accounting for a difference in false match rates.

cs.CV↗

arXiv2021

Multiple Fusion Adaptation: A Strong Framework for Unsupervised Semantic Segmentation Adaptation

This paper challenges the cross-domain semantic segmentation task, aiming to improve the segmentation accuracy on the unlabeled target domain without incurring additional annotation. Using the pseudo-label-based unsupervised domain adaptation (UDA) pipeline, we propose a novel and effective Multiple Fusion Adaptation (MFA) method. MFA basically considers three parallel information fusion strategies, i.e., the cross-model fusion, temporal fusion and a novel online-offline pseudo label fusion. Specifically, the online-offline pseudo label fusion encourages the adaptive training to pay additional attention to difficult regions that are easily ignored by offline pseudo labels, therefore retaining more informative details. While the other two fusion strategies may look standard, MFA pays significant efforts to raise the efficiency and effectiveness for integration, and succeeds in injecting all the three strategies into a unified framework. Experiments on two widely used benchmarks, i.e., GTA5-to-Cityscapes and SYNTHIA-to-Cityscapes, show that our method significantly improves the semantic segmentation adaptation, and sets up new state of the art (58.2% and 62.5% mIoU, respectively). The code will be available at https://github.com/KaiiZhang/MFA.

cs.CV↗

arXiv2021

Multi-Interactive Attention Network for Fine-grained Feature Learning in CTR Prediction

In the Click-Through Rate (CTR) prediction scenario, user's sequential behaviors are well utilized to capture the user interest in the recent literature. However, despite being extensively studied, these sequential methods still suffer from three limitations. First, existing methods mostly utilize attention on the behavior of users, which is not always suitable for CTR prediction, because users often click on new products that are irrelevant to any historical behaviors. Second, in the real scenario, there exist numerous users that have operations a long time ago, but turn relatively inactive in recent times. Thus, it is hard to precisely capture user's current preferences through early behaviors. Third, multiple representations of user's historical behaviors in different feature subspaces are largely ignored. To remedy these issues, we propose a Multi-Interactive Attention Network (MIAN) to comprehensively extract the latent relationship among all kinds of fine-grained features (e.g., gender, age and occupation in user-profile). Specifically, MIAN contains a Multi-Interactive Layer (MIL) that integrates three local interaction modules to capture multiple representations of user preference through sequential behaviors and simultaneously utilize the fine-grained user-specific as well as context information. In addition, we design a Global Interaction Module (GIM) to learn the high-order interactions and balance the different impacts of multiple features. Finally, Offline experiment results from three datasets, together with an Online A/B test in a large-scale recommendation system, demonstrate the effectiveness of our proposed approach.

cs.IR↗

arXiv2021

An IFU View of the Active Galactic Nuclei in MaNGA Galaxy Pairs

The role of active galactic nuclei (AGNs) during galaxy interactions and how they influence the star formation in the system are still under debate. We use a sample of 1156 galaxies in galaxy pairs or mergers (hereafter `pairs') from the MaNGA survey. This pair sample is selected by the velocity offset, projected separation, and morphology, and is further classified into four cases along the merger sequence based on morphological signatures. We then identify a total of 61 (5.5%) AGNs in pairs based on the emission-line diagnostics. No evolution of the AGN fraction is found, either along the merger sequence or compared to isolated galaxies (5.0%). We observe a higher fraction of passive galaxies in galaxy pairs, especially in the pre-merging cases, and associate the higher fraction to their environmental dependence. The isolated AGN and AGN in pairs show similar distributions in their global stellar mass, star formation rate (SFR), and central [OIII] surface brightness. AGNs in pairs show radial profiles of increasing specific SFR and declining Dn4000 from center to outskirts, and no significant difference from the isolated AGNs. This is clearly different from star-forming galaxies (SFGs) in our pair sample, which show enhanced central star formation, as reported before. AGNs in pairs have lower Balmer decrements at outer regions, possibly indicating less dust attenuation. Our findings suggest that AGNs likely follow an inside-out quenching and the merger impact on the star formation in AGNs is less prominent than in SFGs.

astro-ph.GA↗

arXiv2021

A note on the Gagliardo-Nirenberg inequality in a bounded domain

The classical Gagliardo-Nirenberg inequality was established in $\mathbb{R}^n$. An extension to a bounded domain was given by Gagliardo in 1959. In this note, we present a simple proof of this result and prove a new Gagliardo-Nirenberg inequality in a bounded Lipschitz domain.

math.AP↗

arXiv2021

Free Energy Cost to Assemble Superlattices of Polymer-Grafted Nanoparticles

Mesoparticles consisting of a hard core and a soft corona like polymer-grafted nanoparticles (PGNPs) can assemble into various superlattice structures, in which each mesoparticle assumes the shape of the corresponding Wigner-Seitz (or Voronoi) cell. Conventional wisdom often perceives the stability of these superlattices in a mean-field view of surface area minimization or corona entropy maximization, which lacks a molecular interpretation. We develop a simulation method to calculate the free energy cost to deform spherical PGNPs into Wigner-Seitz polyhedra, which are then relaxed in a certain crystalline superlattice. With this method, we successfully quantify the free energy differences between model BCC, FCC and A15 systems of PGNPs and identify BCC as the most stable structure in most cases. Analysis of polymer configurations in the corona, whose boundary is blurred by chain interpenetration, shows that the radial distribution of grafted chains and the corresponding entropy is almost identical among BCC and FCC, suggesting that the higher stability of BCC structure cannot be explained by a mean-field description of corona shape.

cond-mat.soft↗

arXiv2021

Pointwise Boundary Differentiability for Fully Nonlinear Elliptic Equations

In this paper, we prove the pointwise boundary differentiability for viscosity solutions of fully nonlinear elliptic equations. This generalizes the previous related results for linear equations. The geometrical conditions in this paper are pointwise and more general than before. Moreover, our proofs are relatively simple.

math.AP↗

arXiv2021

Generalization techniques of neural networks for fluid flow estimation

We demonstrate several techniques to encourage practical uses of neural networks for fluid flow estimation. In the present paper, three perspectives which are remaining challenges for applications of machine learning to fluid dynamics are considered: 1. interpretability of machine-learned results, 2. bulking out of training data, and 3. generalizability of neural networks. For the interpretability, we first demonstrate two methods to observe the internal procedure of neural networks, i.e., visualization of hidden layers and application of gradient-weighted class activation mapping (Grad-CAM), applied to canonical fluid flow estimation problems -- $(1)$ drag coefficient estimation of a cylinder wake and $(2)$ velocity estimation from particle images. It is exemplified that both approaches can successfully tell us evidences of the great capability of machine learning-based estimations. We then utilize some techniques to bulk out training data for super-resolution analysis and temporal prediction for cylinder wake and NOAA sea surface temperature data to demonstrate that sufficient training of neural networks with limited amount of training data can be achieved for fluid flow problems. The generalizability of machine learning model is also discussed by accounting for the perspectives of inter/extrapolation of training data, considering super-resolution of wakes behind two parallel cylinders. We find that various flow patterns generated by complex interaction between two cylinders can be reconstructed well, even for the test configurations regarding the distance factor. The present paper can be a significant step toward practical uses of neural networks for both laminar and turbulent flow problems.

physics.flu-dyn↗

arXiv2021

Rotation Curves of Galaxies and Their Dependence on Morphology and Stellar Mass

We study how stellar rotation curves (RCs) of galaxies are correlated on average with morphology and stellar mass ($M_\mathrm{star}$) using the final release of Sloan Digital Sky Survey IV MaNGA data. We use the visually assigned $T$-types for the morphology indicator, and adopt a functional form for the RC that can model non-flat RCs at large radii. We discover that within the radial coverage of the MaNGA data, the popularly known flat rotation curve at large radii applies only to the particular classes of galaxies, i.e., massive late types ($T$-type $\geq1$, $M_\mathrm{star}\gtrsim10^{10.8}M_\odot$) and S0 types ($T$-type$=-1$ or $0$, $M_\mathrm{star}\gtrsim10^{10.0}M_\odot$). The RC of late-type galaxies at large radii rises more steeply as $M_\mathrm{star}$ decreases, and its slope increases to about $+9$ km s$^{-1}$kpc$^{-1}$ at $M_\mathrm{star}\approx10^{9.7}M_\odot$. By contrast, elliptical galaxies ($T$-type $\le-2$) have descending RCs at large radii. Their slope becomes more negative as $M_\mathrm{star}$ decreases, and reaches as negative as $-15$ km s$^{-1}$kpc$^{-1}$ at $M_\mathrm{star}\approx10^{10.2}M_\odot$. We also find that the inner slope of the RC is highest for elliptical galaxies with $M_\mathrm{star}\approx10^{10.5}M_\odot$, and decreases as $T$-type increases or $M_\mathrm{star}$ changes away from $10^{10.5}M_\odot$. The velocity at the turnover radius $R_t$ is higher for higher $M_\mathrm{star}$, and $R_t$ is larger for higher $M_\mathrm{star}$ and later $T$-types. We show that the inner slope of the RC is coupled with the central surface stellar mass density, which implies that the gravitational potential of central regions of galaxies is dominated by baryonic matter. With the aid of simple models for matter distribution, we discuss what determines the shapes of RCs.

astro-ph.GA↗

arXiv2021

The AUGUST Two-Sample Test: Powerful, Interpretable, and Fast

Two-sample testing is a fundamental problem in statistics, and many famous two-sample tests are designed to be fully non-parametric. These existing methods perform well with location and scale shifts but are less robust when faced with more exotic classes of alternatives, and rejections from these tests can be difficult to interpret. Here, we propose a new univariate non-parametric two-sample test, AUGUST, designed to improve on these aspects. AUGUST tests for inequality in distribution up to a predetermined resolution using symmetry statistics from binary expansion. The AUGUST statistic is exactly distribution-free and has a well-understood asymptotic distribution, permitting fast p-value computation. In empirical studies, we show that AUGUST has power comparable to that of the best existing methods in every context, as well as greater power in some circumstances. We illustrate the clear interpretability of AUGUST on NBA shooting data.

stat.ME↗

arXiv2021

CondiDiag1.0: A flexible online diagnostic tool for conditional sampling and budget analysis in the E3SM atmosphere model (EAM)

Numerical models used in weather and climate prediction take into account a comprehensive set of atmospheric processes such as the resolved and unresolved fluid dynamics, radiative transfer, cloud and aerosol life cycles, and mass or energy exchanges with the Earth's surface. In order to identify model deficiencies and improve predictive skills, it is important to obtain process-level understanding of the interactions between different processes. Conditional sampling and budget analysis are powerful tools for process-oriented model evaluation, but they often require tedious ad hoc coding and large amounts of instantaneous model output, resulting in inefficient use of human and computing resources. This paper presents an online diagnostic tool that addresses this challenge by monitoring model variables in a generic manner as they evolve within the time integration cycle. The tool is convenient to use. It allows users to select sampling conditions and specify monitored variables at run time. Both the evolving values of the model variables and their increments caused by different atmospheric processes can be monitored and archived. Online calculation of vertical integrals is also supported. Multiple sampling conditions can be monitored in a single simulation in combination with unconditional sampling. The paper explains in detail the design and implementation of the tool in the Energy Exascale Earth System Model (E3SM) version 1. The usage is demonstrated through three examples: a global budget analysis of dust aerosol mass concentration, a composite analysis of sea salt emission and its dependency on surface wind speed, and a conditionally sampled relative humidity budget. The tool is expected to be easily portable to closely related atmospheric models that use the same or similar data structures and time integration methods.

physics.ao-ph↗