arXiv subjects
Publications and source records attributed to Lei Chen.
Bandwidth slicing is introduced to support federated learning in edge computing to assure low communication delay for training traffic. Results reveal that bandwidth slicing significantly improves training efficiency while achieving good learning accuracy.
With broad applications in various public services like aviation management and urban disaster warning, numerical precipitation prediction plays a crucial role in weather forecast. However, constrained by the limitation of observation and conventional meteorological models, the numerical precipitation predictions are often highly biased. To correct this bias, classical correction methods heavily depend on profound experts who have knowledge in aerodynamics, thermodynamics and meteorology. As precipitation can be influenced by countless factors, however, the performances of these expert-driven methods can drop drastically when some un-modeled factors change. To address this issue, this paper presents a data-driven deep learning model which mainly includes two blocks, i.e. a Denoising Autoencoder Block and an Ordinal Regression Block. To the best of our knowledge, it is the first expert-free models for bias correction. The proposed model can effectively correct the numerical precipitation prediction based on 37 basic meteorological data from European Centre for Medium-Range Weather Forecasts (ECMWF). Experiments indicate that compared with several classical machine learning algorithms and deep learning models, our method achieves the best correcting performance and meteorological index, namely the threat scores (TS), obtaining satisfactory visualization effect.
In this paper, we study the action of $\text{Homeo}_0(M)$, the identity component of the group of homeomorphisms of an $n$-dimensional manifold $M$ with an $\mathbb{F}_p$-free action, on another manifold $N$ of dimension $n+k<2n$. We prove that if $M$ is not an $\mathbb{F}_p$-homology sphere, then $N\cong M\times K$ for a homology manifold $K$ such that the action of $\text{Homeo}_0(M)$ on $M$ is standard and on $K$ is trivial. In particular, for $M=S^n$ a sphere, any nontrivial action is a generalization of the "coning-off" construction.
For a closed manifold $M$, let Fib$(M)$ be the number of distinct fiberings of $M$ as a fiber bundle with fiber a closed surface. In this paper we give the first computation of Fib$(M)$ where $1<\text{Fib}(M)<\infty$ but $M$ is not a product. In particular, we prove Fib$(M)=2$ for the Atiyah-Kodaira manifold and any finite cover of a trivial surface bundle. We also give an example where Fib$(M)=4$.
Nielsen realization problem for the mapping class group $\text{Mod}(S_g)$ asks whether the natural projection $p_g: \text{Homeo}_+(S_g)\to \text{Mod}(S_g)$ has a section. While all the previous results use torsion elements in an essential way, in this paper, we focus on the much more difficult problem of realization of torsion-free subgroups of $\text{Mod}(S_g)$. The main result of this paper is that the Torelli group has no realization inside the area-preserving homeomorphisms.
We present ALMA observations of 12CO, 13CO, and C18O J=2--1 lines and the 230 GHz continuum for the FU Ori-type object (FUor) V900 Mon (d~1.5 kpc), for which the accretion burst was triggered between 1953 and 2009. We identified CO emission associated with a molecular bipolar outflow extending up to a ~10^4 au scale and a rotating molecular envelope extending over >10^4 au. The interaction with the hot energetic FUor wind, which was observed using optical spectroscopy, appears limited to a region within ~400 au of the star. The envelope mass and the collimation of the extended CO outflow suggest that the progenitor of this FUor is a low-mass Class I young stellar object (YSO). These parameters for V900 Mon, another FUor, and a few FUor-like stars are consistent with the idea that FUor outbursts are associated with normal YSOs. The continuum emission is marginally resolved in our observations with a 0."2x0."15 (~300x225 au) beam, and a Gaussian model provides a deconvolved FWHM of ~90 au. The emission is presumably associated with a dusty circumstellar disk, plus a possible contribution from a wind or a wind cavity close to the star. The warm compact nature of the disk continuum emission could be explained with viscous heating of the disk, while gravitational fragmentation in the outer disk and/or a combination of grain growth and their inward drift may also contribute to its compact nature.
In this paper, we show that an infinite 2-group of bounded exponent cannot act faithfully and smoothly on compact manifolds.
Lifelong learning is challenging for deep neural networks due to their susceptibility to catastrophic forgetting. Catastrophic forgetting occurs when a trained network is not able to maintain its ability to accomplish previously learned tasks when it is trained to perform new tasks. We study the problem of lifelong learning for generative models, extending a trained network to new conditional generation tasks without forgetting previous tasks, while assuming access to the training data for the current task only. In contrast to state-of-the-art memory replay based approaches which are limited to label-conditioned image generation tasks, a more generic framework for continual learning of generative models under different conditional image generation settings is proposed in this paper. Lifelong GAN employs knowledge distillation to transfer learned knowledge from previous networks to the new network. This makes it possible to perform image-conditioned generation tasks in a lifelong learning setting. We validate Lifelong GAN for both image-conditioned and label-conditioned generation tasks, and provide qualitative and quantitative results to show the generality and effectiveness of our method.
Although the band topology of ZrGeSe has been studied via magnetic torque technique, the electronic transport behaviors related to the relativistic Fermions in ZrGeSe are still unknown. Here, we first report systematic electronic transport properties of high-quality ZrGeSe single crystals under magnetic fields up to 14 T. Resistivity plateaus of temperature dependent resistivity curves both in the presence and absence of magnetic fields as well as large, non-saturating magnetoresistance in low-temperature region were observed. By analyzing the temperature- and angular-dependent Shubnikov-de Haas oscillations and fitting it via the Lifshitz-Kosevich (LK) formula with the Berry phase being taken into account, we proved that Dirac fermions dominate the electronic transport behaviors of ZrGeSe and the presence of non-trivial Berry phase. First principles calculations demonstrate that ZrGeSe possesses Dirac bands and normal bands near Fermi surface, resulting in the observed magnetotransport phenomena. These results demonstrate that ZrGeSe is a topological nodal-line semimetal, which provides a fundamentally important platform to study the quantum physics of topological semimetals.
As personal electronic devices increasingly rely on cloud computing for energy-intensive calculations, the power consumption associated with the information revolution is rapidly becoming an important environmental issue. Several approaches have been proposed to construct electronic devices with low energy consumption. Among these, the low-dissipation surface states of topological insulators (TIs) are widely employed. To develop TI-based devices, a key factor is the maximum temperature at which the Dirac surface states dominate the transport behavior. Here, we employ Shubnikov-de Haas oscillations (SdH) as a means to study the surface state survival temperature in a high quality vanadium doped Bi1.08Sn0.02Sb0.9Te2S single crystal system. The temperature and angle dependence of the SdH show that: 1) crystals with different vanadium (V) doping levels are insulating in the 3-300 K region, 2) the SdH oscillations show two-dimensional behavior, indicating that the oscillations arise from the pure surface states; and 3) at 50 K, the V0.04 single crystals (Vx:Bi1.08-xSn0.02Sb0.9Te2S, where x = 0.04) still show clear sign of SdH oscillations, which demonstrate that the surface dominant transport behavior can survive above 50 K. The robust surface states in our V doped single crystal systems provide an ideal platform to study the Dirac fermions and their interaction with other materials above 50 K.
Legal issues have risen with the changing landscape of computing, especially when the service, data and infrastructure is not owned by the user. With the Cloud, the question arises as to who is in the possession of the data. The Cloud provider can be considered as a legal custodian, owner or possessor of the data thereby causing complexities in legal matters around trademark infringement, privacy of users and their data, abuse and security. By introducing Cloud design focusing on privacy, legal as a service on a Cloud and service provider accountability, users can expect the service providers to be accountable for privacy and data in addition to their regular SLAs.
In this paper we give a close-to-sharp answer to the basic questions: When is there a continuous way to add a point to a configuration of $n$ ordered points on a surface $S$ of finite type so that all the points are still distinct? When this is possible, what are all the ways to do it? More precisely, let PConf$_n(S)$ be the space of ordered $n$-tuple of distinct points in $S$. Let $f_n(S): \text{PConf}_{n+1}(S) \to \text{PConf}_n(S)$ be the map given by $f_n(x_0,x_1,\ldots ,x_n):=(x_1,\ldots ,x_n)$. We classify all continuous sections of $f_n$ up to homotopy by proving the following. 1. If $S=\mathbb{R}^2$ and $n>3$, any section of $f_{n}(S)$ is either "adding a point at infinity" or "adding a point near $x_k$". (We define these two terms in Section 2.1; whether we can define "adding a point near $x_k$" or "adding a point at infinity" depends in a delicate way on properties of $S$. ) 2. If $S=S^2$ a $2$-sphere and $n>4$, any section of $f_{n}(S)$ is "adding a point near $x_k$"; if $S=S^2$ and $n=2$, the bundle $f_n(S)$ does not have a section. (We define this term in Section 3.2) 3. If $S=S_g$ a surface of genus $g>1$ and for $n>1$, we give an easy proof that the bundle $f_{n}(S)$ does not have a section.
In this paper, we will show that the projection $\text{Homeo}^+(D^2_n)\to B_n$ does not have a section; i.e. the braid group $B_n$ cannot be geometrically realized as a group of homeomorphisms of a disk fixing the boundary point-wise and $n$ marked points in the interior as a set. We also give a new proof of a result of Markovic that the mapping class group of a closed surface cannot be geometrically realized as a group of homeomorphisms.
Fizeau experiment, inspiring Einstein's special theory of relativity, reveals a small dragging effect of light inside a moving medium. Dispersion can enhance such light drag according to Lorentz's predication. Here we experimentally demonstrate slow and fast light enhanced light drag in a moving optical microcavity through stimulated Brillouin scattering induced transparency and absorption. The strong dispersion provides an enhancement factor up to ~10^4, greatly reducing the system size down to the micrometer range. These results may offer a unique platform for a compact, integrated solution to motion sensing and ultrafast signal processing applications.
We answer the question of when a new point can be added in a continuous way to configurations of $n$ distinct points in a closed ball of arbitrary dimension. We show that this is possible given an ordered configuration of $n$ points if and only if $n \neq 1$. On the other hand, when the points are not ordered and the dimension of the ball is at least 2, a point can be added continuously if and only if $n = 2$. These results generalize the Brouwer fixed-point theorem, which gives the negative answer when $n=1$. We also show that when $n=2$, there is a unique solution to both the ordered and unordered versions of the problem up to homotopy.
Let $PB_n(S_{g,p})$ be the pure braid group of a genus $g>1$ surface with $p$ punctures. In this paper we prove that any surjective homomorphism $PB_n(S_{g,p})\to PB_m(S_{g,p})$ factors through one of the forgetful homomorphisms. We then compute the automorphism group of $PB_n(S_{g,p})$, extending Irmak, Ivanov and McCarthy's result to the punctured case. Surprisingly, in contrast to the $n=1$ case, any automorphism of $PB_n(S_{g,p})$, $n>1$ is geometric.
Image based social networks are among the most popular social networking services in recent years. With tremendous images uploaded everyday, understanding users' preferences on user-generated images and making recommendations have become an urgent need. In fact, many hybrid models have been proposed to fuse various kinds of side information~(e.g., image visual representation, social network) and user-item historical behavior for enhancing recommendation performance. However, due to the unique characteristics of the user generated images in social image platforms, the previous studies failed to capture the complex aspects that influence users' preferences in a unified framework. Moreover, most of these hybrid models relied on predefined weights in combining different kinds of information, which usually resulted in sub-optimal recommendation performance. To this end, in this paper, we develop a hierarchical attention model for social contextual image recommendation. In addition to basic latent user interest modeling in the popular matrix factorization based recommendation, we identify three key aspects (i.e., upload history, social influence, and owner admiration) that affect each user's latent preferences, where each aspect summarizes a contextual factor from the complex relationships between users and images. After that, we design a hierarchical attention network that naturally mirrors the hierarchical relationship (elements in each aspects level, and the aspect level) of users' latent interests with the identified key aspects. Specifically, by taking embeddings from state-of-the-art deep learning models that are tailored for each kind of data, the hierarchical attention network could learn to attend differently to more or less content. Finally, extensive experimental results on real-world datasets clearly show the superiority of our proposed model.
The anomalous thermodynamic properties of the paradigmatic frustrated spin-1/2 triangular lattice Heisenberg antiferromagnet (TLH) has remained an open topic of research over decades, both experimentally and theoretically. Here we further the theoretical understanding based on the recently developed, powerful exponential tensor renormalization group (XTRG) method on cylinders and stripes in a quasi one-dimensional (1D) setup, as well as a tensor product operator approach directly in 2D. The observed thermal properties of the TLH are in excellent agreement with two recent experimental measurements on the virtually ideal TLH material Ba$_8$CoNb$_6$O$_{24}$. Remarkably, our numerical simulations reveal two crossover temperature scales, at $T_l/J \sim 0.20$ and $T_h/J\sim 0.55$, with $J$ the Heisenberg exchange coupling, which are also confirmed by a more careful inspection of the experimental data. We propose that in the intermediate regime between the low-temperature scale $T_l$ and the higher one $T_h$, the gapped "roton-like" excitations are activated with a strong chiral component and a large contribution to thermal entropies, which suppress the incipient 120$^\circ$ order that emerges for temperatures below $T_l$.