arXiv subjects

Yue Wang

Publications and source records attributed to Yue Wang.

At least 883 records · Page 49Linked to original sources

arXiv2018

Learning-to-Ask: Knowledge Acquisition via 20 Questions

Almost all the knowledge empowered applications rely upon accurate knowledge, which has to be either collected manually with high cost, or extracted automatically with unignorable errors. In this paper, we study 20 Questions, an online interactive game where each question-response pair corresponds to a fact of the target entity, to acquire highly accurate knowledge effectively with nearly zero labor cost. Knowledge acquisition via 20 Questions predominantly presents two challenges to the intelligent agent playing games with human players. The first one is to seek enough information and identify the target entity with as few questions as possible, while the second one is to leverage the remaining questioning opportunities to acquire valuable knowledge effectively, both of which count on good questioning strategies. To address these challenges, we propose the Learning-to-Ask (LA) framework, within which the agent learns smart questioning strategies for information seeking and knowledge acquisition by means of deep reinforcement learning and generalized matrix factorization respectively. In addition, a Bayesian approach to represent knowledge is adopted to ensure robustness to noisy user responses. Simulating experiments on real data show that LA is able to equip the agent with effective questioning strategies, which result in high winning rates and rapid knowledge acquisition. Moreover, the questioning strategies for information seeking and knowledge acquisition boost the performance of each other, allowing the agent to start with a relatively small knowledge set and quickly improve its knowledge base in the absence of constant human supervision.

cs.AI↗

arXiv2018

Maximum Volume Inscribed Ellipsoid: A New Simplex-Structured Matrix Factorization Framework via Facet Enumeration and Convex Optimization

Consider a structured matrix factorization model where one factor is restricted to have its columns lying in the unit simplex. This simplex-structured matrix factorization (SSMF) model and the associated factorization techniques have spurred much interest in research topics over different areas, such as hyperspectral unmixing in remote sensing, topic discovery in machine learning, to name a few. In this paper we develop a new theoretical SSMF framework whose idea is to study a maximum volume ellipsoid inscribed in the convex hull of the data points. This maximum volume inscribed ellipsoid (MVIE) idea has not been attempted in prior literature, and we show a sufficient condition under which the MVIE framework guarantees exact recovery of the factors. The sufficient recovery condition we show for MVIE is much more relaxed than that of separable non-negative matrix factorization (or pure-pixel search); coincidentally it is also identical to that of minimum volume enclosing simplex, which is known to be a powerful SSMF framework for non-separable problem instances. We also show that MVIE can be practically implemented by performing facet enumeration and then by solving a convex optimization problem. The potential of the MVIE framework is illustrated by numerical results.

stat.ML↗

arXiv2018

Topological and Algebraic Properties of Chernoff Information between Gaussian Graphs

In this paper, we want to find out the determining factors of Chernoff information in distinguishing a set of Gaussian graphs. We find that Chernoff information of two Gaussian graphs can be determined by the generalized eigenvalues of their covariance matrices. We find that the unit generalized eigenvalue doesn't affect Chernoff information and its corresponding dimension doesn't provide information for classification purpose. In addition, we can provide a partial ordering using Chernoff information between a series of Gaussian trees connected by independent grafting operations. With the relationship between generalized eigenvalues and Chernoff information, we can do optimal linear dimension reduction with least loss of information for classification.

cs.IT↗

arXiv2018

Code Completion with Neural Attention and Pointer Networks

Intelligent code completion has become an essential research task to accelerate modern software development. To facilitate effective code completion for dynamically-typed programming languages, we apply neural language models by learning from large codebases, and develop a tailored attention mechanism for code completion. However, standard neural language models even with attention mechanism cannot correctly predict the out-of-vocabulary (OoV) words that restrict the code completion performance. In this paper, inspired by the prevalence of locally repeated terms in program source code, and the recently proposed pointer copy mechanism, we propose a pointer mixture network for better predicting OoV words in code completion. Based on the context, the pointer mixture network learns to either generate a within-vocabulary word through an RNN component, or regenerate an OoV word from local context through a pointer component. Experiments on two benchmarked datasets demonstrate the effectiveness of our attention mechanism and pointer mixture network on the code completion task.

cs.CL↗

arXiv2018

Bounded Policy Synthesis for POMDPs with Safe-Reachability Objectives

Planning robust executions under uncertainty is a fundamental challenge for building autonomous robots. Partially Observable Markov Decision Processes (POMDPs) provide a standard framework for modeling uncertainty in many applications. In this work, we study POMDPs with safe-reachability objectives, which require that with a probability above some threshold, a goal state is eventually reached while keeping the probability of visiting unsafe states below some threshold. This POMDP formulation is different from the traditional POMDP models with optimality objectives and we show that in some cases, POMDPs with safe-reachability objectives can provide a better guarantee of both safety and reachability than the existing POMDP models through an example. A key algorithmic problem for POMDPs is policy synthesis, which requires reasoning over a vast space of beliefs (probability distributions). To address this challenge, we introduce the notion of a goal-constrained belief space, which only contains beliefs reachable from the initial belief under desired executions that can achieve the given safe-reachability objective. Our method compactly represents this space over a bounded horizon using symbolic constraints, and employs an incremental Satisfiability Modulo Theories (SMT) solver to efficiently search for a valid policy over it. We evaluate our method using a case study involving a partially observable robotic domain with uncertain obstacles. The results show that our method can synthesize policies over large belief spaces with a small number of SMT solver calls by focusing on the goal-constrained belief space.

cs.RO↗

arXiv2018

Regularity of Solutions of the Camassa-Holm Equations with Fractional Laplacian Viscosity

We study the existence, uniqueness and regularity of solutions to the $n$-dimensional ($n=2,3$) Camassa-Holm equations with fractional Laplacian viscosity with smooth initial data. It is a coupled system between the Navier-Stokes equations with nonlocal viscosity and a Helmholtz equation. The main difficulty lies in establishing some a priori estimates for the fractional Laplacian viscosity. To achieve this, we need to explore suitable fractional-power Sobolev-type estimates, and bilinear estimates for fractional derivatives. Especially, for the critical case $\displaystyle s=\frac{n}{4}$ with $n=2,3$, we will make extra efforts for acquiring the expected estimates as obtained in the case $\displaystyle \frac{n}{4}<s<1$. By the aid of the fractional Leibniz rule and the nonlocal version of Ladyzhenskaya's inequality, we prove the existence, uniqueness and regularity to the Camassa-Holm equations under study by the energy method and a bootstrap argument, which rely crucially on the fractional Laplacian viscosity. In particular, under the critical case $s=\dfrac{n}{4}$, the nonlocal version of Ladyzhenskaya's inequality is skillfully used, and the smallness of initial data in several Sobolev spaces is required to gain the desired results concernig existence, uniqueness and regularity.

math.AP↗

arXiv2018

Im2Avatar: Colorful 3D Reconstruction from a Single Image

Existing works on single-image 3D reconstruction mainly focus on shape recovery. In this work, we study a new problem, that is, simultaneously recovering 3D shape and surface color from a single image, namely "colorful 3D reconstruction". This problem is both challenging and intriguing because the ability to infer textured 3D model from a single image is at the core of visual understanding. Here, we propose an end-to-end trainable framework, Colorful Voxel Network (CVN), to tackle this problem. Conditioned on a single 2D input, CVN learns to decompose shape and surface color information of a 3D object into a 3D shape branch and a surface color branch, respectively. Specifically, for the shape recovery, we generate a shape volume with the state of its voxels indicating occupancy. For the surface color recovery, we combine the strength of appearance hallucination and geometric projection by concurrently learning a regressed color volume and a 2D-to-3D flow volume, which are then fused into a blended color volume. The final textured 3D model is obtained by sampling color from the blended color volume at the positions of occupied voxels in the shape volume. To handle the severe sparse volume representations, a novel loss function, Mean Squared False Cross-Entropy Loss (MSFCEL), is designed. Extensive experiments demonstrate that our approach achieves significant improvement over baselines, and shows great generalization across diverse object categories and arbitrary viewpoints.

cs.CV↗

arXiv2018

Differentially Private Confidence Intervals for Empirical Risk Minimization

The process of data mining with differential privacy produces results that are affected by two types of noise: sampling noise due to data collection and privacy noise that is designed to prevent the reconstruction of sensitive information. In this paper, we consider the problem of designing confidence intervals for the parameters of a variety of differentially private machine learning models. The algorithms can provide confidence intervals that satisfy differential privacy (as well as the more recently proposed concentrated differential privacy) and can be used with existing differentially private mechanisms that train models using objective perturbation and output perturbation.

cs.LG↗

arXiv2018

Some insights into the behaviour of millimetre wave spectrum on key 5G cellular KPIs

This invited paper discusses the challenging task of assessing the millmeter-wave spectrum for the use of 5G mobile systems. 3 key KPIs related to mobile systems, coverage, capacity and mobility are individually assessed and quantified with a simplistic figure of merit (FoM). The coverage and capacity KPIs are assessed through simulations while an analytical approach is developed for mobility. The key message is that the FoM degrades with the higher as the frequencies increase in this range. This indicates that there will be more challenges in deploying millimeter wave systems in higher frequencies, but these are not insurmountable. Many of these challenges are addressed in the mmMAGIC project and possible solutions are derived.

cs.IT↗

arXiv2018

On Overcoming the Impact of Doppler Spectrum in Millimeter-Wave V2I Communications

With the advent of 5G cellular systems there is an increased interest in exploring higher frequency bands above 6 GHz. At these frequencies, beamforming appears as a straightforward solution to overcome higher path loss thereby altering the Doppler characteristics of the received waves. Higher frequencies can suffer from strong Doppler impairments because of the linear dependency of Doppler shift with carrier frequency, which makes them challenging to use in high-mobility scenarios, particularly Vehicular-to-Infrastructure (V2I) communications. Therefore, the impact of beamforming on the Doppler characteristics of the received signals is of utter importance for future V2I systems. This paper presents a theoretical analysis of the Doppler power spectrum in the presence of beamforming at the transmit and/or the receive sides. Further approximations are made for the resulting Doppler spread and Doppler shift when the receive beam width is sufficiently small, and a possible design solution is presented to control the Doppler spread in V2I systems. The results can be of key importance in waveform and air interface design for V2I systems.

cs.IT↗

arXiv2018

Frequency and Quadrature Amplitude Modulation for 5G Networks

The emerging fifth generation (5G) wireless access network, aiming at providing ubiquitous and high data rate connectivity, is envisaged to deploy large number of base stations with higher density and smaller sizes, where inter-cell interference (ICI) becomes a critical problem. Frequency quadrature amplitude modulation (FQAM) has been shown to reduce the ICI at the cell edge therefore achieve a higher transmission rate for cell edge users. This paper investigates the detection of FQAM symbols and noise plus ICI in a multi-cell FQAM communication network. Turbo-coded bit error rate (BER) and frame error rate (FER) of multi-cell FQAM are studied. Also, the cumulative distribution function (CDF) of signal to noise plus interference (SINR) of multi-cell FQAM is computed using stochastic geometry. It is demonstrated via simulations FQAM outperforms quadrature amplitude modulation (QAM) in BER and FER when ICI is significant. Furthermore, FQAM can achieve better SINR than QAM.

cs.IT↗

arXiv2018

Laser map aided visual inertial localization in changing environment

Long-term visual localization in outdoor environment is a challenging problem, especially faced with the cross-seasonal, bi-directional tasks and changing environment. In this paper we propose a novel visual inertial localization framework that localizes against the LiDAR-built map. Based on the geometry information of the laser map, a hybrid bundle adjustment framework is proposed, which estimates the poses of the cameras with respect to the prior laser map as well as optimizes the state variables of the online visual inertial odometry system simultaneously. For more accurate cross-modal data association, the laser map is optimized using multi-session laser and visual data to extract the salient and stable subset for localization. To validate the efficiency of the proposed method, we collect data in south part of our campus in different seasons, along the same and opposite-direction route. In all sessions of localization data, our proposed method gives satisfactory results, and shows the superiority of the hybrid bundle adjustment and map optimization.

cs.RO↗

arXiv2018

3D-SSD: Learning Hierarchical Features from RGB-D Images for Amodal 3D Object Detection

This paper aims at developing a faster and a more accurate solution to the amodal 3D object detection problem for indoor scenes. It is achieved through a novel neural network that takes a pair of RGB-D images as the input and delivers oriented 3D bounding boxes as the output. The network, named 3D-SSD, composed of two parts: hierarchical feature fusion and multi-layer prediction. The hierarchical feature fusion combines appearance and geometric features from RGB-D images while the multi-layer prediction utilizes multi-scale features for object detection. As a result, the network can exploit 2.5D representations in a synergetic way to improve the accuracy and efficiency. The issue of object sizes is addressed by attaching a set of 3D anchor boxes with varying sizes to every location of the prediction layers. At the end stage, the category scores for 3D anchor boxes are generated with adjusted positions, sizes and orientations respectively, leading to the final detections using non-maximum suppression. In the training phase, the positive samples are identified with the aid of 2D ground truth to avoid the noisy estimation of depth from raw data, which guide to a better converged model. Experiments performed on the challenging SUN RGB-D dataset show that our algorithm outperforms the state-of-the-art Deep Sliding Shape by 10.2% mAP and 88x faster. Further, experiments also suggest our approach achieves comparable accuracy and is 386x faster than the state-of-art method on the NYUv2 dataset even with a smaller input image size.

cs.CV↗

arXiv2018

Energy-aware Adaptive Spectrum Access and Power Allocation in LAA Networks via Lyapunov Optimization

To relieve the traffic burden and improve the system capacity, licensed-assisted access (LAA) has been becoming a promising technology to the supplementary utilization of the unlicensed spectrum. However, due to the densification of small base stations (SBSs) and the dynamic variety of the number of Wi-Fi nodes in the overlapping areas, the licensed channel interference and the unlicensed channel collision could seriously influence the Quality of Service (QoS) and the energy consumption. In this paper, jointly considering time-variant wireless channel conditions, dynamic traffic loads, and random numbers of Wi-Fi nodes, we address an adaptive spectrum access and power allocation problem that enables minimizing the system power consumption under a certain queue stability constraint in the LAA-enabled SBSs and Wi-Fi networks. The complex stochastic optimization problem is rewritten as the difference of two convex (D.C.) program in the framework of Lyapunov optimization, thus developing an online energy-aware optimal algorithm. We also characterize the performance bounds of the proposed algorithm with a tradeoff of [O(1=V ); O(V )] between power consumption and delay theoretically. The numerical results verify the tradeoff and show that our scheme can reduce the power consumption over the existing scheme by up to 72.1% under the same traffic delay.

eess.SP↗

arXiv2017

Chernoff Information between Gaussian Trees

In this paper, we aim to provide a systematic study of the relationship between Chernoff information and topological, as well as algebraic properties of the corresponding Gaussian tree graphs for the underlying graphical testing problems. We first show the relationship between Chernoff information and generalized eigenvalues of the associated covariance matrices. It is then proved that Chernoff information between two Gaussian trees sharing certain local subtree structures can be transformed into that of two smaller trees. Under our proposed grafting operations, bottleneck Gaussian trees, namely, Gaussian trees connected by one such operation, can thus be simplified into two 3-node Gaussian trees, whose topologies and edge weights are subject to the specifics of the operation. Thereafter, we provide a thorough study about how Chernoff information changes when small differences are accumulated into bigger ones via concatenated grafting operations. It is shown that the two Gaussian trees connected by more than one grafting operation may not have bigger Chernoff information than that of one grafting operation unless these grafting operations are separate and independent. At the end, we propose an optimal linear dimensional reduction method related to generalized eigenvalues.

cs.IT↗

arXiv2017

Revisiting the electronic phase diagram of YBa2Cu3Oy via temperature derivative of in-plane resistivity

We have re-examined the temperature-doping (T-p) phase diagram of YBa2Cu3Oy (YBCO) to address several issues therein by using the temperature derivative of in-plane resistivity, d\{rho}ab/dT. For p less than about 0.15, a temperature Tf has been defined to mark the onset of an upturn in d\{rho}ab/dT at T>Tc, which, in light of prior studies on another cuprate La2-xSrxCuO4, is attributed to the onset of superconducting fluctuations in normal state. The Tf exhibits a doping dependence similar to Tc, and the interval between Tf and Tc is about 5-30 K across the underdoped regime, showing agreement with a variety of other measurements to probe a restricted temperature range of superconducting fluctuations. Above Tf, the d\{rho}ab/dT increases linearly as T increases up to a value denoted as T2, which is about half the T* and falls roughly in parallel with T* as p rises up to about 0.13, indicating a prominent T^2-dependent \{rho}ab in this T-p region. The d\{rho}ab/dT helps reveal that, at Tf 0.10-0.11, thereby leaving a narrow window of doping for \{rho}ab to show a pure T^2 dependence. As T increases further, the d\{rho}ab/dT reaches a local maximum at a temperature T_R, signifying the known curvature change (inflection point) in \{rho}ab. With the derivatives, it is illustrated that, in the vicinity of T_R, the in-plane Hall coefficient R_H and thermopower Sab of YBCO display curvature changes as well, suggesting a correlation of the three transport properties. It is also found that the dSab/dT shows the onset of an upturn at a temperature coinciding with the Tf, which, backing the identification of Tf with the onset of superconducting fluctuations, demonstrates further the virtue of using the temperature derivative to unveil information helpful for the study of high-Tc cuprates.

cond-mat.supr-con↗

arXiv2017

1305 nm MoTe2-on-silicon Laser

The missing piece in the jigsaw of silicon photonics is a light source that can be easily incorporated into the standard silicon fabrication process. Recent advances in the development of atomically thin layers of semiconducting transition metal dichalogenides (TMDs), with direct bandgaps in the near-infrared region, have opened up new possibilities for addressing this need. Here, we report a unique silicon laser source that employs molybdenum ditelluride (MoTe2) as a gain material in a photonic crystal nanocavity resonator, fabricated in silicon-on-insulator. We demonstrate optically pumped MoTe2-on-silicon devices lasing at 1305 nm, i.e. in the centre of the O-band used in optical communications, operating in the continuous-wave (CW) regime, at room temperature and with a threshold power density as low as 1.5 kW/cm2. This 2D-on-silicon geometry offers the promise of an integrated low-cost electrically pumped nanoscale silicon light source, thereby adding an essential building block to the silicon photonics platform.

cond-mat.mes-hall↗

arXiv2017

Convergence Analysis of Distributed Stochastic Gradient Descent with Shuffling

When using stochastic gradient descent to solve large-scale machine learning problems, a common practice of data processing is to shuffle the training data, partition the data across multiple machines if needed, and then perform several epochs of training on the re-shuffled (either locally or globally) data. The above procedure makes the instances used to compute the gradients no longer independently sampled from the training data set. Then does the distributed SGD method have desirable convergence properties in this practical situation? In this paper, we give answers to this question. First, we give a mathematical formulation for the practical data processing procedure in distributed machine learning, which we call data partition with global/local shuffling. We observe that global shuffling is equivalent to without-replacement sampling if the shuffling operations are independent. We prove that SGD with global shuffling has convergence guarantee in both convex and non-convex cases. An interesting finding is that, the non-convex tasks like deep learning are more suitable to apply shuffling comparing to the convex tasks. Second, we conduct the convergence analysis for SGD with local shuffling. The convergence rate for local shuffling is slower than that for global shuffling, since it will lose some information if there's no communication between partitioned data. Finally, we consider the situation when the permutation after shuffling is not uniformly distributed (insufficient shuffling), and discuss the condition under which this insufficiency will not influence the convergence rate. Our theoretical results provide important insights to large-scale machine learning, especially in the selection of data processing methods in order to achieve faster convergence and good speedup. Our theoretical findings are verified by extensive experiments on logistic regression and deep neural networks.

stat.ML↗