arXiv subjects
Publications and source records attributed to Guan Huang.
In this paper, we show that any smooth one-parameter deformations of a strictly convex integrable billiard table $Ω_0$ preserving the integrability near the boundary have to be tangent to a finite dimensional space passing through $Ω_0$.
In order to mitigate the long processing delay and high energy consumption of mobile augmented reality (AR) applications, mobile edge computing (MEC) has been recently proposed and is envisioned as a promising means to deliver better quality of experience (QoE) for AR consumers. In this article, we first present a comprehensive AR overview, including the indispensable components of general AR applications, fashionable AR devices, and several existing techniques for overcoming the thorny latency and energy consumption problems. Then, we propose a novel hierarchical computation architecture by inserting an edge layer between the conventional user layer and cloud layer. Based on the proposed architecture, we further develop an innovated operation mechanism to improve the performance of mobile AR applications. Three key technologies are also discussed to further assist the proposed AR architecture. Simulation results are finally provided to verify that our proposals can significantly improve the latency and energy performance as compared against existing baseline schemes.
Estimating post-click conversion rate (CVR) accurately is crucial for ranking systems in industrial applications such as recommendation and advertising. Conventional CVR modeling applies popular deep learning methods and achieves state-of-the-art performance. However it encounters several task-specific problems in practice, making CVR modeling challenging. For example, conventional CVR models are trained with samples of clicked impressions while utilized to make inference on the entire space with samples of all impressions. This causes a sample selection bias problem. Besides, there exists an extreme data sparsity problem, making the model fitting rather difficult. In this paper, we model CVR in a brand-new perspective by making good use of sequential pattern of user actions, i.e., impression -> click -> conversion. The proposed Entire Space Multi-task Model (ESMM) can eliminate the two problems simultaneously by i) modeling CVR directly over the entire space, ii) employing a feature representation transfer learning strategy. Experiments on dataset gathered from Taobao's recommender system demonstrate that ESMM significantly outperforms competitive methods. We also release a sampling version of this dataset to enable future research. To the best of our knowledge, this is the first public dataset which contains samples with sequential dependence of click and conversion labels for CVR modeling.
The Birkhoff conjecture says that the boundary of a strictly convex integrable billiard table is necessarily an ellipse. In this article, we consider a stronger notion of integrability, namely integrability close to the boundary, and prove a local version of this conjecture: a small perturbation of an ellipse of small eccentricity which preserves integrability near the boundary, is itself an ellipse. This extends the result in [1], where integrability was assumed on a larger set. In particular, it shows that (local) integrability near the boundary implies global integrability. One of the crucial ideas in the proof consists in analyzing Taylor expansion of the corresponding action-angle coordinates with respect to the eccentricity parameter, deriving and studying higher order conditions for the preservation of integrable rational caustics.
Convolutional neural networks (CNN) based tracking approaches have shown favorable performance in recent benchmarks. Nonetheless, the chosen CNN features are always pre-trained in different task and individual components in tracking systems are learned separately, thus the achieved tracking performance may be suboptimal. Besides, most of these trackers are not designed towards real-time applications because of their time-consuming feature extraction and complex optimization details.In this paper, we propose an end-to-end framework to learn the convolutional features and perform the tracking process simultaneously, namely, a unified convolutional tracker (UCT). Specifically, The UCT treats feature extractor and tracking process both as convolution operation and trains them jointly, enabling learned CNN features are tightly coupled to tracking process. In online tracking, an efficient updating method is proposed by introducing peak-versus-noise ratio (PNR) criterion, and scale changes are handled efficiently by incorporating a scale branch into network. The proposed approach results in superior tracking performance, while maintaining real-time speed. The standard UCT and UCT-Lite can track generic objects at 41 FPS and 154 FPS without further optimization, respectively. Experiments are performed on four challenging benchmark tracking datasets: OTB2013, OTB2015, VOT2014 and VOT2015, and our method achieves state-of-the-art results on these benchmarks compared with other real-time trackers.
In this paper we show that for a generic strictly convex domain, one can recover the eigendata corresponding to Aubry-Mather periodic orbits of the induced billiard map, from the (maximal) marked length spectrum of the domain.
Consider weakly nonlinear complex Ginzburg--Landau (CGL) equation of the form: $$ u_t+i(-Δu+V(x)u)=εμΔu+ε\mathcal{P}( u),\quad x\in {R^d}\,, \quad(*) $$ under the periodic boundary conditions, where $μ\geqslant0$ and $\mathcal{P}$ is a smooth function. Let $\{ζ_1(x),ζ_2(x),\dots\}$ be the $L_2$-basis formed by eigenfunctions of the operator $-Δ+V(x)$. For a complex function $u(x)$, write it as $u(x)=\sum_{k\geqslant1}v_kζ_k(x)$ and set $I_k(u)=\frac{1}{2}|v_k|^2$. Then for any solution $u(t,x)$ of the linear equation $(*)_{ε=0}$ we have $I(u(t,\cdot))=const$. In this work it is proved that if equation $(*)$ with a sufficiently smooth real potential $V(x)$ is well posed on time-intervals $t\lesssim ε^{-1}$, then for any its solution $u^ε(t,x)$, the limiting behavior of the curve $I(u^ε(t,\cdot))$ on time intervals of order $ε^{-1}$, as $ε\to0$, can be uniquely characterized by a solution of a certain well-posed effective equation: $$ u_t=εμ\triangle u+εF(u), $$ where $F(u)$ is a resonant averaging of the nonlinearity $\mathcal{P}(u)$. We also prove a similar results for the stochastically perturbed equation, when a white in time and smooth in $x$ random force of order $\sqrtε$ is added to the right-hand side of the equation. The approach of this work is rather general. In particular, it applies to equations in bounded domains in $R^d$ under Dirichlet boundary conditions.
To date, there have been massive Semi-Structured Documents (SSDs) during the evolution of the Internet. These SSDs contain both unstructured features (e.g., plain text) and metadata (e.g., tags). Most previous works focused on modeling the unstructured text, and recently, some other methods have been proposed to model the unstructured text with specific tags. To build a general model for SSDs remains an important problem in terms of both model fitness and efficiency. We propose a novel method to model the SSDs by a so-called Tag-Weighted Topic Model (TWTM). TWTM is a framework that leverages both the tags and words information, not only to learn the document-topic and topic-word distributions, but also to infer the tag-topic distributions for text mining tasks. We present an efficient variational inference method with an EM algorithm for estimating the model parameters. Meanwhile, we propose three large-scale solutions for our model under the MapReduce distributed computing platform for modeling large-scale SSDs. The experimental results show the effectiveness, efficiency and the robustness by comparing our model with the state-of-the-art methods in document modeling, tags prediction and text classification. We also show the performance of the three distributed solutions in terms of time and accuracy on document modeling.
Consider a weakly nonlinear CGL equation on the torus~$\mathbb{T}^d$: \[u_t+iΔu=ε[μ(-1)^{m-1}Δ^{m} u+b|u|^{2p}u+ ic|u|^{2q}u].\eqno{(*)}\] Here $u=u(t,x)$, $x\in\mathbb{T}^d$, $0<ε<<1$, $μ\geqslant0$, $b,c\in\mathbb{R}$ and $m,p,q\in\mathbb{N}$. Define \mbox{$I(u)=(I_{\dk},\dk\in\mathbb{Z}^d)$}, where $I_{\dk}=v_{\dk}\bar{v}_{\dk}/2$ and $v_{\dk}$, $\dk\in\mathbb{Z}^d$, are the Fourier coefficients of the function~$u$ we give. Assume that the equation $(*)$ is well posed on time intervals of order $ε^{-1}$ and its solutions have there a-priori bounds, independent of the small parameter. Let $u(t,x)$ solve the equation $(*)$. If $ε$ is small enough, then for $t\lesssimε^{-1}$, the quantity $I(u(t,x))$ can be well described by solutions of an {\it effective equation}: \[u_t=ε[μ(-1)^{m-1}Δ^m u+ F(u)],\] where the term $F(u)$ can be constructed through a kind of resonant averaging of the nonlinearity $b|u|^{2p}+ ic|u|^{2q}u$.
Consider perturbed KdV equations: \[u_t+u_{xxx}-6uu_x=εf(u(\cdot)),\quad x\in\mathbb{T}=\mathbb{R}/\mathbb{Z},\;\int_{\mathbb{T}}u(x,t)dx=0,\] where the nonlinearity defines analytic operators $u(\cdot)\mapsto f(u(\cdot))$ in sufficiently smooth Sobolev spaces. Assume that the equation has an $ε$-quasi-invariant measure $μ$ and satisfies some additional mild assumptions. Let $u^ε(t)$ be a solution. Then on time intervals of order $ε^{-1}$, as $ε\to0$, its actions $I(u^ε(t,\cdot))$ can be approximated by solutions of a certain well-posed averaged equation, provided that the initial datum is $μ$-typical.
Consider nonlinear Schrödinger equations with small nonlinearities \[\frac{d}{dt}u+i(-\triangle u+V(x)u)=ε\mathcal{P}(\triangle u,u,x),\quad x\in \mathbb{T}^d.\eqno{(*)}\] Let $\{ζ_1(x),ζ_2(x),\dots\}$ be the $L_2$-basis formed by eigenfunctions of the operator $-\triangle +V(x)$. For any complex function $u(x)$, write it as \mbox{$u(x)=\sum_{k\geqslant1}v_kζ_k(x)$} and set $I_k(u)=\frac{1}{2}|v_k|^2$. Then for any solution $u(t,x)$ of the linear equation $(*)_{ε=0}$ we have $I(u(t,\cdot))=const$. In this work it is proved that if $(*)$ is well posed on time-intervals $t\lesssim ε^{-1}$ and satisfies there some mild a-priori assumptions, then for any its solution $u^ε(t,x)$, the limiting behavior of the curve $I(u^ε(t,\cdot))$ on time intervals of order $ε^{-1}$, as $ε\to0$, can be uniquely characterized by solutions of a certain well-posed effective equation.
In this paper we discuss properties of the KdV equation under periodic boundary conditions, especially those which are important to study perturbations of the equation. Next we review what is known now about long-time behaviour of solutions for perturbed KdV equations.
We consider a perturbed KdV equation: [\dot{u}+u_{xxx} - 6uu_x = εf(x,u(\cdot)), \quad x\in \mathbb{T}, \quad\int_\mathbb{T} u dx=0.] For any periodic function $u(x)$, let $I(u)=(I_1(u),I_2(u),...)\in\mathbb{R}_+^{\infty}$ be the vector, formed by the KdV integrals of motion, calculated for the potential $u(x)$. Assuming that the perturbation $εf(x,u(\cdot))$ is a smoothing mapping (e.g. it is a smooth function $εf(x)$, independent from $u$), and that solutions of the perturbed equation satisfy some mild a-priori assumptions, we prove that for solutions $u(t,x)$ with typical initial data and for $0\leqslant t\lesssim ε^{-1}$, the vector $I(u(t))$ may be well approximated by a solution of the averaged equation.