arXiv subjects

Hui Wu

Publications and source records attributed to Hui Wu.

At least 73 records · Page 4Linked to original sources

arXiv2019

Drill-down: Interactive Retrieval of Complex Scenes using Natural Language Queries

This paper explores the task of interactive image retrieval using natural language queries, where a user progressively provides input queries to refine a set of retrieval results. Moreover, our work explores this problem in the context of complex image scenes containing multiple objects. We propose Drill-down, an effective framework for encoding multiple queries with an efficient compact state representation that significantly extends current methods for single-round image retrieval. We show that using multiple rounds of natural language queries as input can be surprisingly effective to find arbitrarily specific images of complex scenes. Furthermore, we find that existing image datasets with textual captions can provide a surprisingly effective form of weak supervision for this task. We compare our method with existing sequential encoding and embedding networks, demonstrating superior performance on two proposed benchmarks: automatic image retrieval on a simulated scenario that uses region captions as queries, and interactive image retrieval using real queries from human evaluators.

cs.CV↗

arXiv2019

Learn to Sense: a Meta-learning Based Sensing and Fusion Framework for Wireless Sensor Networks

Wireless sensor networks (WSN) acts as the backbone of Internet of Things (IoT) technology. In WSN, field sensing and fusion are the most commonly seen problems, which involve collecting and processing of a huge volume of spatial samples in an unknown field to reconstruct the field or extract its features. One of the major concerns is how to reduce the communication overhead and data redundancy with prescribed fusion accuracy. In this paper, an integrated communication and computation framework based on meta-learning is proposed to enable adaptive field sensing and reconstruction. It consists of a stochastic-gradient-descent (SGD) based base-learner used for the field model prediction aiming to minimize the average prediction error, and a reinforcement meta-learner aiming to optimize the sensing decision by simultaneously rewarding the error reduction with samples obtained so far and penalizing the corresponding communication cost. An adaptive sensing algorithm based on the above two-layer meta-learning framework is presented. It actively determines the next most informative sensing location, and thus considerably reduces the spatial samples and yields superior performance and robustness compared with conventional schemes. The convergence behavior of the proposed algorithm is also comprehensively analyzed and simulated. The results reveal that the proposed field sensing algorithm significantly improves the convergence rate.

eess.SP↗

arXiv2018

Dialog-based Interactive Image Retrieval

Existing methods for interactive image retrieval have demonstrated the merit of integrating user feedback, improving retrieval results. However, most current systems rely on restricted forms of user feedback, such as binary relevance responses, or feedback based on a fixed set of relative attributes, which limits their impact. In this paper, we introduce a new approach to interactive image search that enables users to provide feedback via natural language, allowing for more natural and effective interaction. We formulate the task of dialog-based interactive image retrieval as a reinforcement learning problem, and reward the dialog system for improving the rank of the target image during each dialog turn. To mitigate the cumbersome and costly process of collecting human-machine conversations as the dialog system learns, we train our system with a user simulator, which is itself trained to describe the differences between target and candidate images. The efficacy of our approach is demonstrated in a footwear retrieval application. Experiments on both simulated and real-world data show that 1) our proposed learning framework achieves better accuracy than other supervised and reinforcement learning baselines and 2) user feedback based on natural language rather than pre-specified attributes leads to more effective retrieval results, and a more natural and expressive communication interface.

cs.CV↗

arXiv2018

Towards the growth of single-crystal boron nitride monolayer on Cu

Atom-layered hexagonal boron nitride (hBN), with excellent stability, flat surface and large bandgap, has been reported to be the best 2D insulator to open up the great possibilities for exciting potential applications in electronics, optoelectronics and photovoltaics. The ability to grow high-quality large single crystals of hBN is at the heart for those applications, but the size of single-crystal 2D BN is less than a millimetre till now. Here, we report the first epitaxial growth of a 10*10 cm2 single-crystal hBN monolayer on a low symmetry Cu(110) "vicinal surface". The growth kinetics, unidirectional alignment and seamless stitching of hBN domains are unambiguously illustrated using centimetre- to the atomic-scale characterization techniques. The findings in this work are expected to significantly boost the massive applications of 2D materials-based devices, and also pave the way for the epitaxial growth of broad non-centrosymmetric 2D materials.

cond-mat.mtrl-sci↗

arXiv2018

Dynamics of Kuramoto oscillators with time-delayed positive and negative couplings

Many real-world examples of distributed oscillators involve not only time delays but also attractive (positive) and repulsive (negative) influences in their network interactions. Here, considering such examples, we generalize the Kuramoto model of globally coupled oscillators with time-delayed positive and negative couplings to explore the effects of such couplings in collective phase synchronization. We analytically derive the exact solutions for stable incoherent and coherent states in terms of the system parameters allowing us to precisely understand the interplay of time delays and couplings in collective synchronization. Dependent on these parameters, fully coherent, incoherent states and mixed states are possible. Time-delays especially in the negative coupling seem to facilitate collective synchronization. In case of a stronger negative coupling than positive one, a stable synchronized state cannot be achieved without time delays. We discuss the implications of the model and the results for natural systems, particularly neuronal network systems in the brain.

nlin.AO↗

arXiv2018

Exact explosive synchronization transitions in Kuramoto oscillators with time-delayed coupling

Synchronization commonly occurs in many natural and man-made systems, from neurons in the brain to cardiac cells to power grids to Josephson junction arrays. Transitions to or out of synchrony for coupled oscillators depend on several factors, such as individual frequencies, coupling, interaction time delays and network structure-function relation. Here, using a generalized Kuramoto model of time-delay coupled phase oscillators with frequency-weighted coupling, we study the stability of incoherent and coherent states and the transitions to or out of explosive (abrupt, first-order like) phase synchronization. We analytically derive the exact formulas for the critical coupling strengths at different time delays in both directions of increasing (forward) and decreasing (backward) coupling strengths. We find that time-delay does not affect the transition for the backward direction but shifts the transition for the forward direction of increasing coupling strength. These results provide valuable insights into our understanding of dynamical mechanisms for explosive synchronization in presence of often unavoidable time delays present in many physical and biological systems.

nlin.AO↗

arXiv2018

A Systematic Analysis for State-of-the-Art 3D Lung Nodule Proposals Generation

Lung nodule proposals generation is the primary step of lung nodule detection and has received much attention in recent years . In this paper, we first construct a model of 3-dimension Convolutional Neural Network (3D CNN) to generate lung nodule proposals, which can achieve the state-of-the-art performance. Then, we analyze a series of key problems concerning the training performance and efficiency. Firstly, we train the 3D CNN model with data in different resolutions and find out that models trained by high resolution input data achieve better lung nodule proposals generation performances especially for nodules in too small sizes, while consumes much more memory at the same time. Then, we analyze the memory consumptions on different platforms and the experimental results indicate that CPU architecture can provide us with larger memory and enables us to explore more possibilities of 3D applications. We implement the 3D CNN model on CPU platform and propose an Intel Extended-Caffe framework which supports many highly-efficient 3D computations, which is opened source at https://github.com/extendedcaffe/extended-caffe.

cs.CV↗

arXiv2017

Depth Control of Model-Free AUVs via Reinforcement Learning

In this paper, we consider depth control problems of an autonomous underwater vehicle (AUV) for tracking the desired depth trajectories. Due to the unknown dynamical model of the AUV, the problems cannot be solved by most of model-based controllers. To this purpose, we formulate the depth control problems of the AUV as continuous-state, continuous-action Markov decision processes (MDPs) under unknown transition probabilities. Based on deterministic policy gradient (DPG) and neural network approximation, we propose a model-free reinforcement learning (RL) algorithm that learns a state-feedback controller from sampled trajectories of the AUV. To improve the performance of the RL algorithm, we further propose a batch-learning scheme through replaying previous prioritized trajectories. We illustrate with simulations that our model-free method is even comparable to the model-based controllers as LQI and NMPC. Moreover, we validate the effectiveness of the proposed RL algorithm on a seafloor data set sampled from the South China Sea.

cs.RO↗

arXiv2016

S3Pool: Pooling with Stochastic Spatial Sampling

Feature pooling layers (e.g., max pooling) in convolutional neural networks (CNNs) serve the dual purpose of providing increasingly abstract representations as well as yielding computational savings in subsequent convolutional layers. We view the pooling operation in CNNs as a two-step procedure: first, a pooling window (e.g., $2\times 2$) slides over the feature map with stride one which leaves the spatial resolution intact, and second, downsampling is performed by selecting one pixel from each non-overlapping pooling window in an often uniform and deterministic (e.g., top-left) manner. Our starting point in this work is the observation that this regularly spaced downsampling arising from non-overlapping windows, although intuitive from a signal processing perspective (which has the goal of signal reconstruction), is not necessarily optimal for \emph{learning} (where the goal is to generalize). We study this aspect and propose a novel pooling strategy with stochastic spatial sampling (S3Pool), where the regular downsampling is replaced by a more general stochastic version. We observe that this general stochasticity acts as a strong regularizer, and can also be seen as doing implicit data augmentation by introducing distortions in the feature maps. We further introduce a mechanism to control the amount of distortion to suit different datasets and architectures. To demonstrate the effectiveness of the proposed approach, we perform extensive experiments on several popular image classification benchmarks, observing excellent improvements over baseline models. Experimental code is available at https://github.com/Shuangfei/s3pool.

cs.LG↗

arXiv2016

A Mott insulator continuously connected to iron pnictide superconductors

Iron-based superconductivity develops near an antiferromagnetic order and out of a bad metal normal state, which has been interpreted as originating from a proximate Mott transition. Whether an actual Mott insulator can be realized in the phase diagram of the iron pnictides remains an open question. Here we use transport, transmission electron microscopy, X-ray absorption spectroscopy, and neutron scattering to demonstrate that NaFe$_{1-x}$Cu$_x$As near $x\approx 0.5$ exhibits real space Fe and Cu ordering, and are antiferromagnetic insulators with the insulating behavior persisting above the Néel temperature, indicative of a Mott insulator. Upon decreasing $x$ from $0.5$, the antiferromagnetic ordered moment continuously decreases, yielding to superconductivity around $x=0.05$. Our discovery of a Mott insulating state in NaFe$_{1-x}$Cu$_x$As thus makes it the only known Fe-based material in which superconductivity can be smoothly connected to the Mott insulating state, highlighting the important role of electron correlations in the high-$T_{\rm c}$ superconductivity.

cond-mat.supr-con↗

arXiv2016

Two new parent compounds for FeSe-based superconducting phases

It is well established that the occurrence of superconductivity in iron pnictides is closely related to the tetragonal to orthorhombic structural and antiferromagnetic (AFM) phase transitions. It, however, has not been clear whether the same scenario is appropriate for iron chalcogenide counterparts due to the absence of parent compounds for the latter family of superconductors. Here, we report the synthesis and structure determination of two novel phases in ethylenediamine intercalated FeSe, one is tetragonal and the other orthorhombic in room temperature, which can be stabilized with neutral spacer layers. Both phases can be regarded as the parent compounds for superconductivity as they are non-superconducting (non-SC) in pristine form and superconducting (SC) with Tc up to 38K and 46K, respectively after Na doping, and the switch between SC and no-SC is reversible. Moreover, the two non-SC parent compounds show no evidence of long-range magnetic ordering down to 2K, only with dynamic spin fluctuations at low temperatures, suggesting that no competition between SC and AFM ordering. Our results reveal that undoped iron selenides are quantum paramagnetic in ground state, implying that they are distinct from their pnictide counterparts in pairing mechanism.

cond-mat.supr-con↗

arXiv2016

Antiferromagnetic and Orbital Ordering on a Diamond Lattice Near Quantum Criticality

We present neutron scattering measurements on powder samples of the spinel FeSc2S4 that reveal a previously unobserved magnetic ordering transition occurring at 11.8(2)~K. Magnetic ordering occurs subsequent to a subtle cubic-to-tetragonal structural transition which distorts Fe coordinating sulfur tetrahedra lifting the orbital degeneracy. The application of 1~GPa hydrostatic pressure appears to destabilize this Néel state, reducing the transition temperature to 8.6(8)~K and redistributing magnetic spectral weight to higher energies. The relative magnitudes of ordered $\langle m \rangle^2\!=\!3.1(2)$ and fluctuating moments $\langle δm \rangle^2\!=\!13(1)$ show that the magnetically ordered ground state of FeSc2S4 is drastically renormalized and in proximity to criticality.

cond-mat.str-el↗

arXiv2015

Virtual-MIMO-Boosted Information Propagation on Highways

In vehicular communications, traffic-related information should be spread over the network as quickly as possible to maintain a safer transportation system. This motivates us to develop more efficient information propagation schemes. In this paper, we propose a novel virtual-MIMO-enabled information dissemination scheme, in which the vehicles opportunistically form virtual antenna arrays to boost the transmission range and therefore accelerate information propagation along the highway. We model the information propagation process as a renewal reward process and investigate in detail the \emph{Information Propagation Speed} (IPS) of the proposed scheme. The corresponding closed-form IPS is derived, which shows that the IPS increases cubically with the vehicle density but will ultimately converge to a constant upper bound. Moreover, increased mobility also facilitates the information spreading by offering more communication opportunities. However, the limited network density essentially determines the bottleneck in information spreading. Extensive simulations are carried out to verify our analysis. We also show that the proposed scheme exhibits a significant IPS gain over its conventional counterpart.

cs.IT↗

arXiv2014

Faster Information Propagation on Highways: a Virtual MIMO Approach

In vehicular communications, traffic-related information should be spread over the network as quickly as possible to maintain a safe and reliable transportation system. This motivates us to develop more efficient information propagation schemes. In this paper, we propose a novel cluster-based cooperative information forwarding scheme, in which the vehicles opportunistically form virtual antenna arrays to boost one-hop transmission range and therefore accelerate information propagation along the highway. Both closed-form results of the transmission range gain and the improved Information Propagation Speed (IPS) are derived and verified by simulations. It is observed that the proposed scheme demonstrates the most significant IPS gain in moderate traffic scenarios, whereas too dense or too sparse vehicle density results in less gain. Moreover, it is also shown that increased mobility offers more contact opportunities and thus facilitates information propagation.

cs.IT↗

arXiv2014

Spatial control of surface plasmon polariton excitation at planar metal surface

We illustrate that the surface plasmon polariton (SPP) excitation through the prism coupling method is fundamentally limited by destructive interference of spatial light components. We propose that the destructive interference can be canceled out by tailoring the relative phase for the different spatial components. As a numerical demonstration, we show that through the phase modulation the excited SPP field is concentrated to a hot energy spot, and the SPP field intensity is dramatically enhanced about three folds in comparison with a conventional Gaussian beam illumination.

physics.optics↗

arXiv2013

Structure and Properties of α-NaFeO2-type Ternary Sodium Iridates

The synthesis, structure, and elementary magnetic and electronic properties are reported for layered compounds of the type Na3-xMIr2O6 and Na3-xM2IrO6, where M is a transition metal from the 3d series (M=Zn, Cu, Ni, Co, Fe and Mn). The rhombohedral structures, in space group R-3m, were determined by refinement of neutron and synchrotron powder diffraction data. No clear evidence for long range 2:1 or 1:2 honeycomb-like M/Ir ordering was found in the neutron powder diffraction patterns except in the case of M = Zn, thus in general the compounds are best designated as sodium deficient α-NaFeO2-type phases with formulas Na1-xM1/3Ir2/3O2 or Na1-xM2/3Ir1/3O2. Synchrotron powder diffraction patterns indicate that several of the compounds likely have honeycomb in-plane metal-iridium ordering with disordered stacking of the layers. All the compounds are sodium deficient under our synthetic conditions and are black and insulating. Weiss constants derived from magnetic susceptibility measurements indicate that Na0.62Mn0.61Ir0.39O2, Na0.80Fe2/3Ir1/3O2, Na0.92Ni1/3Ir2/3O2, Na0.86Cu1/3Ir2/3O2, and Na0.89Zn1/3Ir2/3O2 display dominant antiferromagnetic interactions. For Na0.90Co1/3Ir2/3O2 the dominant magnetic interactions at low temperature are ferromagnetic while at high temperatures they are antiferromagnetic; there is also a change in the effective moment. Low temperature specific heat measurements (to 2 K) on Na0.92Ni1/3Ir2/3O2 indicate the presence of a broad magnetic ordering transition. X-ray absorption spectroscopy shows that iridium is at or close to the 4+ oxidation state in all compounds. 23Na nuclear magnetic resonance measurements comparing Na2IrO3 to Na0.92Ni1/3Ir2/3O2 and Na0.89Zn1/3Ir2/3O2 provide strong indications that the electron spins are short-range ordered in the latter two materials. All of the compounds are spin glasses.

cond-mat.mtrl-sci↗

arXiv2013

k-DAG Based Lifetime Aware Data Collection in Wireless Sensor Networks

Wireless Sensor Networks need to be organized for efficient data collection and lifetime maximization. In this paper, we propose a novel routing structure, namely k-DAG, to balance the load of the base station's neighbours while providing the worst-case latency guarantee for data collection, and a distributed algorithm for construction a k-DAG based on a SPD (Shortest Path DAG). In a k-DAG, the lengths of the longest path and the shortest path of each sensor node to the base station differ by at most k. By adding sibling edges to a SPD, our distributed algorithm allows critical nodes to have more routing choices. The simulation results show that our approach significantly outperforms the SPD-based data collection approach in both network lifetime and load balance.

cs.NI↗

arXiv2013

Sustainable high critical temperature in a hydrocarbon superconductor

Organic superconductors are unique materials with a crystal structure made primarily of a complex carbon based network, an element associated directly with life, which were postulated to have a high critical temperature, $T_{C}$, even above room temperature, from a theoretical viewpoint. Pressure plays an essential role in the study of superconductivity in such organic materials, including creation of the first organic superconductor as well as the achievement of the highest $T_{C}$ of 14.2 K for charge transfer salts and 38 K for metal-doped fullerides. However, superconductivity in these organic systems is only sustainable within a very narrow pressure range (a few GPa) and is readily destroyed upon further compression. Here we report high-pressure magnetic susceptibility and structure measurements on a newly discovered superconductor, LaPhenanthrene. It is found that the application of pressure not only significantly increases $T_{C}$ from its ambient-pressure value of 4.8 K to 12.3 K at 18.4 GPa but also stabilizes the high $T_{C}$ over the entire pressure range (61 GPa) of the study. The evolution of $T_{C}$ with pressure is closely correlated with the angle $β$ of the monoclinic unit cell. Both $T_{C}$ and $β$ change sharply with increasing pressure initially but remain constant above 40 GPa, indicating that molecule orientation is essential to superconductivity. These behaviors can be understood in terms of pressure tuning of electron correlations, illustrating the unconventional nature of superconductivity in these hydrocarbon superconductors.

cond-mat.supr-con↗