arXiv subjects

Hui Li

Publications and source records attributed to Hui Li.

At least 631 records · Page 35Linked to original sources

arXiv2020

Knowledge-Augmented Dexterous Grasping with Incomplete Sensing

Humans can determine a proper strategy to grasp an object according to the measured physical attributes or the prior knowledge of the object. This paper proposes an approach to determining the strategy of dexterous grasping by using an anthropomorphic robotic hand simply based on a label or a description of an object. Object attributes are parsed from natural-language descriptions and augmented with an object knowledge base that is scraped from retailer websites. A novel metric named joint probability distance is defined to measure distance between object attributes. The probability distribution of grasp types for the given object is learned using a deep neural network which takes in object features as input. The action of the multi-fingered hand with redundant degrees of freedom (DoF) is controlled by a linear inverse-kinematics model of grasp topology and scales. The grasping strategy generated by the proposed approach is evaluated both by simulation and execution on a Sawyer robot with an AR10 robotic hand.

cs.RO↗

arXiv2020

Stability of the Stokes Immersed Boundary Problem with Bending and Stretching Energy

We study the motion of a 1-D closed elastic string with bending and stretching energy immersed in a 2-D Stokes flow. In this paper we introduce the curve's tangent angle function and the stretching function to describe the deferent deformations of the elastic string. These two functions are defined on the arc-length coordinate and the material coordinate respectively. With the help of the fundamental solution of the Stokes equation, we reformulate the problem into a parabolic system which is called the contour dynamic system. Under the non-self-intersecting and well-stretched assumptions on initial configurations, we establish the local well-posedness of the free boundary problem in Sobolev space. When the initial configurations are sufficiently close to the equilibrium state (i.e. an evenly parametrized circle), we prove that the solutions can be extended globally and the global solutions will converge to the equilibrium state exponentially as t $\to$ +$\infty$.

math.AP↗

arXiv2020

Monitoring-based Differential Privacy Mechanism Against Query-Flooding Parameter Duplication Attack

Public intelligent services enabled by machine learning algorithms are vulnerable to model extraction attacks that can steal confidential information of the learning models through public queries. Though there are some protection options such as differential privacy (DP) and monitoring, which are considered promising techniques to mitigate this attack, we still find that the vulnerability persists. In this paper, we propose an adaptive query-flooding parameter duplication (QPD) attack. The adversary can infer the model information with black-box access and no prior knowledge of any model parameters or training data via QPD. We also develop a defense strategy using DP called monitoring-based DP (MDP) against this new attack. In MDP, we first propose a novel real-time model extraction status assessment scheme called Monitor to evaluate the situation of the model. Then, we design a method to guide the differential privacy budget allocation called APBA adaptively. Finally, all DP-based defenses with MDP could dynamically adjust the amount of noise added in the model response according to the result from Monitor and effectively defends the QPD attack. Furthermore, we thoroughly evaluate and compare the QPD attack and MDP defense performance on real-world models with DP and monitoring protection.

cs.CR↗

arXiv2020

The effects of subgrid models on the properties of giant molecular clouds in galaxy formation simulations

Recent cosmological hydrodynamical simulations are able to reproduce numerous statistical properties of galaxies that are consistent with observational data. Yet, the adopted subgrid models strongly affect the simulation outcomes, limiting the predictive power of these simulations. In this work, we perform a suite of isolated galactic disk simulations under the {\it SMUGGLE} framework and investigate how different subgrid models affect the properties of giant molecular clouds (GMCs). We employ {\sc astrodendro}, a hierarchical clump-finding algorithm, to identify GMCs in the simulations. We find that different choices of subgrid star formation efficiency, $ε_{\rm ff}$, and stellar feedback channels, yield dramatically different mass and spatial distributions for the GMC populations. Without feedback, the mass function of GMCs has a shallower power-law slope and extends to higher mass ranges compared to runs with feedback. Moreover, higher $ε_{\rm ff}$ results in faster molecular gas consumption and steeper mass function slopes. Feedback also suppresses power in the two-point correlation function (TPCF) of the spatial distribution of GMCs. Specifically, radiative feedback strongly reduces the TPCF on scales below 0.2~kpc, while supernova feedback reduces power on scales above 0.2~kpc. Finally, runs with higher $ε_{\rm ff}$ exhibit a higher TPCF than runs with lower $ε_{\rm ff}$, because the dense gas is depleted more efficiently thereby facilitating the formation of well-structured supernova bubbles. We argue that comparing simulated and observed GMC populations can help better constrain subgrid models in the next-generation of galaxy formation simulations.

astro-ph.GA↗

arXiv2020

Thermomagnetic instability of plasma composition gradients

We show that, under Braginskii magneto-hydrodynamics, anti-parallel gradients in average ion charge state and electron temperature can be unstable to the growth of self-generated magnetic fields. The instability is analogous to the field-generating thermomagnetic instability, although it is driven by the collisional thermal force magnetic source term rather than the Biermann battery term. The gradient in ion charge state causes a gradient in collisionality, which couples with temperature perturbations to create a self-generated magnetic field. This magnetic field deflects the electron heat flux in a way that reinforces the temperature perturbation. The derived linearized growth rate, typically on hydrodynamic timescales, includes the resistive and thermal smoothing. It increases with large ion composition gradients and electron heat flux, conditions typical of the hohlraum walls or contaminant mix jets in inertial confinement fusion implosions. However, extended magneto-hydrodynamic simulations indicate that the instability is usually dominated and stabilized by the nonlinear Nernst advection, in a similar manner to the standard thermomagnetic instability.

physics.plasm-ph↗

arXiv2020

Fermi-type particle acceleration from magnetic reconnection at the termination shock of a relativistic striped wind

An oblique-rotating pulsar generates a relativistic striped wind in a pulsar wind nebula (PWN). The termination shock of the PWN compresses the Poynting-flux-dominated flow and drives magnetic reconnection. By carrying out particle-in-cell (PIC) simulations of the termination shock of the PWN, we study the shock structure as well as the energy conversion processes and particle acceleration mechanisms. With the recent advances in the numerical methods, we extend the simulations to the ultra-relativistic regime with bulk Lorentz factor up to γ_{0}=10^{6}. Magnetic reconnection at the termination shock is highly efficient at converting magnetic energy to particle kinetic energy and accelerating particles to high energies. We find that the resulting energy spectra crucially depend on λ/d_{e}. When λ/d_{e} is large (λ\gtrsim40d_{e}) , the downstream particle spectra form a power-law distribution in the magnetically dominated relativistic wind regime with upstream magnetization parameter σ_{0}=10. By analyzing particle trajectories and statistical quantities relevant to particle energization, we find that Fermi-type mechanism dominates the particle acceleration and power-law formation. We find that the results for particle acceleration are scalable as γ_{0} and σ_{0} increase to large values. The maximum energy for electrons and positrons can reach hundreds of TeV if the wind has a bulk Lorentz factor γ_{0}\approx10^{6} and magnetization parameter σ_{0}=10, which can explain the recent observations of high-energy gamma-rays from pulsar wind nebulae (PWNe).

astro-ph.HE↗

arXiv2020

Where to Look and How to Describe: Fashion Image Retrieval with an Attentional Heterogeneous Bilinear Network

Fashion products typically feature in compositions of a variety of styles at different clothing parts. In order to distinguish images of different fashion products, we need to extract both appearance (i.e., "how to describe") and localization (i.e.,"where to look") information, and their interactions. To this end, we propose a biologically inspired framework for image-based fashion product retrieval, which mimics the hypothesized twostream visual processing system of human brain. The proposed attentional heterogeneous bilinear network (AHBN) consists of two branches: a deep CNN branch to extract fine-grained appearance attributes and a fully convolutional branch to extract landmark localization information. A joint channel-wise attention mechanism is further applied to the extracted heterogeneous features to focus on important channels, followed by a compact bilinear pooling layer to model the interaction of the two streams. Our proposed framework achieves satisfactory performance on three image-based fashion product retrieval benchmarks.

cs.CV↗

arXiv2020

Deep Learning is Singular, and That's Good

In singular models, the optimal set of parameters forms an analytic set with singularities and classical statistical inference cannot be applied to such models. This is significant for deep learning as neural networks are singular and thus "dividing" by the determinant of the Hessian or employing the Laplace approximation are not appropriate. Despite its potential for addressing fundamental issues in deep learning, singular learning theory appears to have made little inroads into the developing canon of deep learning theory. Via a mix of theory and experiment, we present an invitation to singular learning theory as a vehicle for understanding deep learning and suggest important future work to make singular learning theory directly applicable to how deep learning is performed in practice.

cs.LG↗

arXiv2020

Symmetric set of transport coefficients for collisional magnetized plasma

In Braginskii extended magneto-hydrodynamics (ExMHD), applicable to collisional astrophysical and high energy density plasmas, the electric field and heat flow are described by the $α$, $β$ and $κ$ transport coefficients. We show that magnetic transport relies primarily on $β_\parallel-β_\perp$ and $α_\perp-α_\parallel$, rather than $α_\perp$ and $β_\perp$ themselves. However, commonly used coefficient fit functions [Epperlein and Haines, Phys. Fluids 29, 1029 (1986)] cannot accurately calculate these quantities. This means that many ExMHD simulations have significantly over-estimated the cross-gradient Nernst advection, resulting in artificial magnetic dissipation and discontinuities. We repeat the kinetic analysis to provide fits that rectify this problem. Use of these in the Gorgon ExMHD code resolves the known discrepancies with kinetic simulations in the literature. Recognizing the fundamental importance of $α_\perp-α_\parallel$ and $β_\parallel-β_\perp$, we re-cast the set of coefficients to find that each of them now shares the same underlying properties. This makes explicit the symmetry of the magnetic and thermal transport equations, as well as the symmetry of the coefficients themselves.

physics.plasm-ph↗

arXiv2020

Fast and Secure Distributed Nonnegative Matrix Factorization

Nonnegative matrix factorization (NMF) has been successfully applied in several data mining tasks. Recently, there is an increasing interest in the acceleration of NMF, due to its high cost on large matrices. On the other hand, the privacy issue of NMF over federated data is worthy of attention, since NMF is prevalently applied in image and text analysis which may involve leveraging privacy data (e.g, medical image and record) across several parties (e.g., hospitals). In this paper, we study the acceleration and security problems of distributed NMF. Firstly, we propose a distributed sketched alternating nonnegative least squares (DSANLS) framework for NMF, which utilizes a matrix sketching technique to reduce the size of nonnegative least squares subproblems with a convergence guarantee. For the second problem, we show that DSANLS with modification can be adapted to the security setting, but only for one or limited iterations. Consequently, we propose four efficient distributed NMF methods in both synchronous and asynchronous settings with a security guarantee. We conduct extensive experiments on several real datasets to show the superiority of our proposed methods. The implementation of our methods is available at https://github.com/qianyuqiu79/DSANLS.

cs.LG↗

arXiv2020

Give Me Something to Eat: Referring Expression Comprehension with Commonsense Knowledge

Conventional referring expression comprehension (REF) assumes people to query something from an image by describing its visual appearance and spatial location, but in practice, we often ask for an object by describing its affordance or other non-visual attributes, especially when we do not have a precise target. For example, sometimes we say 'Give me something to eat'. In this case, we need to use commonsense knowledge to identify the objects in the image. Unfortunately, these is no existing referring expression dataset reflecting this requirement, not to mention a model to tackle this challenge. In this paper, we collect a new referring expression dataset, called KB-Ref, containing 43k expressions on 16k images. In KB-Ref, to answer each expression (detect the target object referred by the expression), at least one piece of commonsense knowledge must be required. We then test state-of-the-art (SoTA) REF models on KB-Ref, finding that all of them present a large drop compared to their outstanding performance on general REF datasets. We also present an expression conditioned image and fact attention (ECIFA) network that extract information from correlated image regions and commonsense knowledge facts. Our method leads to a significant improvement over SoTA REF models, although there is still a gap between this strong baseline and human performance. The dataset and baseline models will be released.

cs.CV↗

arXiv2020

Single Cell Transcriptome Research in Human Placenta

Human placenta is a complex and heterogeneous organ interfacing between the mother and the fetus that supports fetal development. Alterations to placental structural components are associated with various pregnancy complications. To reveal the heterogeneity among various placenta cell types in normal and diseased placentas, as well as elucidate molecular interactions within a population of placental cells, a new genomics technology called single cell RNA-Seq (or scRNA-seq) has been employed in the last couple of years. Here we review the principles of scRNA-seq technology, and summarize the recent human placenta studies at scRNA-seq level across gestational ages as well as in pregnancy complications such as preterm birth and preeclampsia. We list the computational analysis platforms and resources available for the public use. Lastly, we discuss the future areas of interest for placenta single cell studies, as well as the data analytics needed to accomplish them.

q-bio.TO↗

arXiv2020

Retention of Long-Period Gas Giant Planets: Type II Migration Revisited

During their formation, emerging protoplanets tidally interact with their natal disks. Proto-gas-giant planets, with Hills radius larger than the disk thickness, open gaps and quench gas flow in the vicinity of their orbits. It is usually assumed that their type II migration is coupled to the viscous evolution of the disk. Although this hypothesis provides an explanation for the origin of close-in planets, it also encounter predicament on the retention of long-period orbits for most gas giant planets. Moreover, numerical simulations indicate that planets migrations are not solely determined by the viscous diffusion of their natal disk. Here we carry out a series of hydrodynamic simulations combined with analytic studies to examine the transition between different paradigms of type II migration. We find a range of planetary mass for which gas continues to flow through a severely depleted gap so that the surface density distribution in the disk region beyond the gap is maintained in a quasi-steady state. The associated gap profile modifies the location of corotation \& Lindblad resonances. In the proximity of the planet's orbit, high-order Lindblad \& corotation torque are weakened by the gas depletion in the gap while low-order Lindblad torques near the gap walls preserves their magnitude. Consequently, the intrinsic surface density distribution of the disk determines delicately both pace and direction of planets' type II migration. We show that this effect might stall the inward migration of giant planets and preserve them in disk regions where the surface density is steep.

astro-ph.EP↗

arXiv2020

Attacking Recommender Systems with Augmented User Profiles

Recommendation Systems (RS) have become an essential part of many online services. Due to its pivotal role in guiding customers towards purchasing, there is a natural motivation for unscrupulous parties to spoof RS for profits. In this paper, we study the shilling attack: a subsistent and profitable attack where an adversarial party injects a number of user profiles to promote or demote a target item. Conventional shilling attack models are based on simple heuristics that can be easily detected, or directly adopt adversarial attack methods without a special design for RS. Moreover, the study on the attack impact on deep learning based RS is missing in the literature, making the effects of shilling attack against real RS doubtful. We present a novel Augmented Shilling Attack framework (AUSH) and implement it with the idea of Generative Adversarial Network. AUSH is capable of tailoring attacks against RS according to budget and complex attack goals, such as targeting a specific user group. We experimentally show that the attack impact of AUSH is noticeable on a wide range of RS including both classic and modern deep learning based RS, while it is virtually undetectable by the state-of-the-art attack detection model.

cs.IR↗

arXiv2020

Thermodynamical Evolution of Supra-Arcade Downflows

Supra-arcade downflows (SADs) are dark, teardrop-shaped features descending upon flare arcades. They are thought to be the results of magnetic reconnection, but the detailed formation mechanism of SADs and their relationship to flare energy release are still unclear. In this work, we explore the thermodynamical properties of SADs in the 2011 October 22 limb flare using the observations of the Solar Dynamics Observatory/Atmospheric Imaging Assembly and an improved Differential Emission Measure (DEM) technique. Different heating events around SADs are identified and the propagation speeds of plasma heating are calculated. The first heating event starts with the increase of the emission measure at temperatures higher than 5 MK, about 2.8 minutes before the arrival of the first studied SAD. Its propagation speed is about 140 km/s, a little faster than the speed of the SAD. However, the other two events have fast propagation speeds more than 700 km/s. We suspect that the first heating event can be explained by adiabatic compression, but the others may have different causes. Besides, we observed that SADs can push away their surrounding spikes. The formation of SADs is further explained on the basis of patchy and bursty magnetic reconnection that reconnection outflows may push away surrounding plasma and leave dark lanes behind them. The reliability of the DEM results, heating and cooling mechanisms, and other SAD explanations are discussed.

astro-ph.SR↗

arXiv2020

Radiative Hydrodynamic Simulations of the Spectral Characteristics of Solar White-light Flares

As one of the most violent activities in the solar atmosphere, white-light flares (WLFs) are generally known for their enhanced white-light (or continuum) emission, which primarily originates in the solar lower atmosphere. However, we know little about how white-light emission is produced. In this study, we aim to investigate the response of the continua at 3600Å and 4250Å and also the H$α$ and Ly$α$ lines during WLFs modeled with radiative hydrodynamics simulations. We take non-thermal electron beams as the energy source for the WLFs in two different initial atmospheres and vary their parameters. Our results show that the model with non-thermal electron beam heating can clearly show enhancements in the continua at 3600Å and 4250Å as well as in the H$α$ and Ly$α$ lines. A larger electron beam flux, a smaller spectral index, or a penumbral initial atmosphere leads to a stronger emission increase at 3600Å, 4250Å and in the H$α$ line. For the Ly$α$ line, however, it is more preferably enhanced in a quiet-Sun initial atmosphere with a larger spectral index of the electron beam. It is also notable that the continua at 3600Å and 4250Å and the H$α$ line exhibit a dimming at the beginning of the heating and reach their peak emissions later than the peak time of the heating function, while the Ly$α$ line does not show such behaviors. These results can be served as a reference for analyzing future WLF observations.

astro-ph.SR↗

arXiv2020

Magnetic field generation from composition gradients in inertial confinement fusion fuel

Experimental asymmetries in fusion implosions can lead to magnetic field generation in the hot plasma core. For typical parameters, previous studies found that the magnetisation Hall parameter, given by the product of the electron gyro-frequency and Coulomb collision time, can exceed one. This will affect the hydrodynamics through inhibition and deflection of the electron heat flux. The magnetic field source is the collisionless Biermann term, which arises from the Debye shielding potential in electron pressure gradients. We show that there is an additional source term due to the Z dependence of the Coulomb collision operator. If there are ion composition gradients, such as jets of carbon ablator mix entering the hot-spot, this source term can rapidly exceed the Biermann fields. In addition, the Biermann fields are enhanced due to the increased temperature gradients from carbon radiative cooling. With even stronger self-generated fields, heat loss to the carbon regions will be reduced, potentially reducing the negative effect of carbon mix.

physics.plasm-ph↗

arXiv2020

NestFuse: An Infrared and Visible Image Fusion Architecture based on Nest Connection and Spatial/Channel Attention Models

In this paper we propose a novel method for infrared and visible image fusion where we develop nest connection-based network and spatial/channel attention models. The nest connection-based network can preserve significant amounts of information from input data in a multi-scale perspective. The approach comprises three key elements: encoder, fusion strategy and decoder respectively. In our proposed fusion strategy, spatial attention models and channel attention models are developed that describe the importance of each spatial position and of each channel with deep features. Firstly, the source images are fed into the encoder to extract multi-scale deep features. The novel fusion strategy is then developed to fuse these features for each scale. Finally, the fused image is reconstructed by the nest connection-based decoder. Experiments are performed on publicly available datasets. These exhibit that our proposed approach has better fusion performance than other state-of-the-art methods. This claim is justified through both subjective and objective evaluation. The code of our fusion method is available at https://github.com/hli1221/imagefusion-nestfuse

cs.CV↗