Searcharxiv⌕ Search

arXiv subjects

Jun Li

Publications and source records attributed to Jun Li.

At least 739 records · Page 41Linked to original sources
arXiv2020

Efficient Riemannian Optimization on the Stiefel Manifold via the Cayley Transform

Strictly enforcing orthonormality constraints on parameter matrices has been shown advantageous in deep learning. This amounts to Riemannian optimization on the Stiefel manifold, which, however, is computationally expensive. To address this challenge, we present two main contributions: (1) A new efficient retraction map based on an iterative Cayley transform for optimization updates, and (2) An implicit vector transport mechanism based on the combination of a projection of the momentum and the Cayley transform on the Stiefel manifold. We specify two new optimization algorithms: Cayley SGD with momentum, and Cayley ADAM on the Stiefel manifold. Convergence of Cayley SGD is theoretically analyzed. Our experiments for CNN training demonstrate that both algorithms: (a) Use less running time per iteration relative to existing approaches that enforce orthonormality of CNN parameters; and (b) Achieve faster convergence rates than the baseline SGD and ADAM algorithms without compromising the performance of the CNN. Cayley SGD and Cayley ADAM are also shown to reduce the training time for optimizing the unitary transition matrices in RNNs.

cs.LG↗
arXiv2020

Learning Part Generation and Assembly for Structure-aware Shape Synthesis

Learning powerful deep generative models for 3D shape synthesis is largely hindered by the difficulty in ensuring plausibility encompassing correct topology and reasonable geometry. Indeed, learning the distribution of plausible 3D shapes seems a daunting task for the holistic approaches, given the significant topological variations of 3D objects even within the same category. Enlightened by the fact that 3D shape structure is characterized as part composition and placement, we propose to model 3D shape variations with a part-aware deep generative network, coined as PAGENet. The network is composed of an array of per-part VAE-GANs, generating semantic parts composing a complete shape, followed by a part assembly module that estimates a transformation for each part to correlate and assemble them into a plausible structure. Through delegating the learning of part composition and part placement into separate networks, the difficulty of modeling structural variations of 3D shapes is greatly reduced. We demonstrate through both qualitative and quantitative evaluations that PAGENet generates 3D shapes with plausible, diverse and detailed structure, and show two applications, i.e., semantic shape segmentation and part-based shape editing.

cs.CV↗
arXiv2020

Quantum Pure State Tomography via Variational Hybrid Quantum-Classical Method

To obtain a complete description of a quantum system, one usually employs standard quantum state tomography, which however requires exponential number of measurements to perform and hence is impractical when the system's size grows large. In this work, we introduce a self-learning tomographic scheme based on the variational hybrid quantum-classical method. The key part of the scheme is a learning procedure, in which we learn a control sequence capable of driving the unknown target state coherently to a simple fiducial state, so that the target state can be directly reconstructed by applying the control sequence reversely. In this manner, the state tomography problem is converted to a state-to-state transfer problem. To solve the latter problem, we use the closed-loop learning control approach. Our scheme is further experimentally tested using techniques of a 4-qubit nuclear magnetic resonance. {Experimental results indicate that the proposed tomographic scheme can handle a broad class of states including entangled states in quantum information, as well as dynamical states of quantum many-body systems common to condensed matter physics.

quant-ph↗
arXiv2020

Experimental Detection of the Quantum Phases of a Three-Dimensional Topological Insulator on a Spin Quantum Simulator

The detection of topological phases of matter becomes a central issue in recent years. Conventionally, the realization of a specific topological phase in condensed matter physics relies on probing the underlying surface band dispersion or quantum transport signature of a real material, which may be imperfect or even absent. On the other hand, quantum simulation offers an alternative approach to directly measure the topological invariant on a universal quantum computer. However, experimentally demonstrating high-dimensional topological phases remains a challenge due to the technical limitations of current experimental platforms. Here, we investigate the three-dimensional topological insulators in the AIII (chiral unitary) symmetry class which yet lack experimental realization. Using the nuclear magnetic resonance system, we experimentally demonstrate their topological properties, where a dynamical quenching approach is adopted and the dynamical bulk-boundary correspondence in the momentum space is observed. As a result, the topological invariants are measured with high precision on the band-inversion surface, exhibiting robustness to the decoherence effect. Our work paves the way towards the quantum simulation of topological phases of matter in higher dimensions and more complex systems through controllable quantum phases transitions.

quant-ph↗
arXiv2020

UAV-Enabled Confidential Data Collection in Wireless Networks

This work, for the first time, considers confidential data collection in the context of unmanned aerial vehicle (UAV) wireless networks, where the scheduled ground sensor node (SN) intends to transmit confidential information to the UAV without being intercepted by other unscheduled ground SNs. Specifically, a full-duplex (FD) UAV collects data from each scheduled SN on the ground and generates artificial noise (AN) to prevent the scheduled SN's confidential information from being wiretapped by other unscheduled SNs. We first derive the reliability outage probability (ROP) and secrecy outage probability (SOP) of a considered fixed-rate transmission, based on which we formulate an optimization problem that maximizes the minimum average secrecy rate (ASR) subject to some specific constraints. We then transform the formulated optimization problem into a convex problem with the aid of first-order restrictive approximation technique and penalty method. The resultant problem is a generalized nonlinear convex programming (GNCP) and solving it directly still leads to a high complexity, which motivates us to further approximate this problem as a second-order cone program (SOCP) in order to reduce the computational complexity. Finally, we develop an iteration procedure based on penalty successive convex approximation (P-SCA) algorithm to pursue the solution to the formulated optimization problem. Our examination shows that the developed joint design achieves a significant performance gain compared to a benchmark scheme.

eess.SP↗
arXiv2019

Colliding Winds in and around the Stellar Group IRS 13E at the Galactic Center

IRS~13E is an enigmatic compact group of massive stars located in projection only 3.6 arcseconds away from Sgr A*. This group has been suggested to be bounded by an intermediate-mass black hole (IMBH). We present a multi-wavelength study of the group and its interplay with the environment. Based on Chandra observations, we find the X-ray spectrum of IRS~13E can be well characterized by an optically thin thermal plasma. The emission peaks between two strongly mass-losing Wolf-Rayet stars of the group. These properties can be reasonably well reproduced by simulated colliding winds of these two stars. However, this scenario under-predicts the X-ray intensity in outer regions. The residual emission likely results from the ram-pressure confinement of the IRS~13E group wind by the ambient medium and is apparently associated with a shell-like warm gas structure seen in Pa-alpha and in ALMA observations. These latter observations also show strongly peaked thermal emission with unusually large velocity spread between the two stars. These results indicate that the group is colliding with the bar of the dense cool gas mini-spiral around Sgr A*. The extended X-ray morphology of IRS~13E and its association with the bar further suggest that the group is physically much farther away than the projected distance from Sgr A*. The presence of an IMBH, while favorable to keep the stars bound together, is not necessary to explain the observed stellar and gas properties of IRS~13E.

astro-ph.HE↗
arXiv2019

Experimental Observation of Equilibrium and Dynamical Quantum Phase Transitions via Out-of-Time-Ordered Correlators

The out-of-time-ordered correlators (OTOC) have been established as a fundamental concept for quantifying quantum information scrambling and diagnosing quantum chaotic behavior. Recently, it was theoretically proposed that the OTOC can be used as an order parameter to dynamically detect both equilibrium quantum phase transitions (EQPTs) and dynamical quantum phase transitions (DQPTs) in one-dimensional many-body systems. Here we report the first experimental observation of EQPTs and DQPTs in a quantum spin chain via quench dynamics of OTOC on a nuclear magnetic resonance quantum simulator. We observe that the quench dynamics of both the order parameter and the two-body correlation function cannot detect the DQPTs, but the OTOC can unambiguously detect the DQPTs. Moreover, we demonstrate that the long-time average value of the OTOC in quantum quench signals the equilibrium quantum critical point and ordered quantum phases, thus one can measure the EQPTs from the non-equilibrium quantum quench dynamics. Our experiment paves a way for experimentally investigating DQPTs through OTOCs and for studying the EQPTs through the non-equilibrium quantum quench dynamics with quantum simulators.

quant-ph↗
arXiv2019

Lifelong Spectral Clustering

In the past decades, spectral clustering (SC) has become one of the most effective clustering algorithms. However, most previous studies focus on spectral clustering tasks with a fixed task set, which cannot incorporate with a new spectral clustering task without accessing to previously learned tasks. In this paper, we aim to explore the problem of spectral clustering in a lifelong machine learning framework, i.e., Lifelong Spectral Clustering (L2SC). Its goal is to efficiently learn a model for a new spectral clustering task by selectively transferring previously accumulated experience from knowledge library. Specifically, the knowledge library of L2SC contains two components: 1) orthogonal basis library: capturing latent cluster centers among the clusters in each pair of tasks; 2) feature embedding library: embedding the feature manifold information shared among multiple related tasks. As a new spectral clustering task arrives, L2SC firstly transfers knowledge from both basis library and feature library to obtain encoding matrix, and further redefines the library base over time to maximize performance across all the clustering tasks. Meanwhile, a general online update formulation is derived to alternatively update the basis library and feature library. Finally, the empirical experiments on several real-world benchmark datasets demonstrate that our L2SC model can effectively improve the clustering performance when comparing with other state-of-the-art spectral clustering algorithms.

cs.LG↗
arXiv2019

Symplectic $(-2)$ spheres and the symplectomorphism group of small rational 4-manifolds, I

Let $(X,ω)$ be a symplectic rational 4 manifold. We study the space of tamed almost complex structures $\mathcal{J}_ω$ using a fine decomposition via smooth rational curves and a relative version of the infinite-dimensional Alexander duality. This decomposition provides new understandings of both the variation and stability of the symplectomorphism group $Symp(X,ω)$ when deforming $ω$. In particular, we compute the rank of $π_1(Symp(X,ω))$ with Euler number $χ(X)\leq 7$ in terms of the number $N$ of -2 symplectic sphere classes.

math.SG↗
arXiv2019

Symplectic $(-2)$-spheres and the symplectomorphism group of small rational 4-manifolds, II

For $(\mathbb{C} P^2 \# 5{\overline {\mathbb{C} P^2}},ω)$, let $N_ω$ be the number of $(-2)$-symplectic spherical homology classes.We completely determine the Torelli symplectic mapping class group (Torelli SMCG): the Torelli SMCG is trivial if $N_ω>8$; it is $π_0(Diff^+(S^2,5))$ if $N_ω=0$ (by Paul Seidel and Jonathan Evans); it is $π_0(Diff^+(S^2,4))$ in the remaining case. Further, we completely determine the rank of $π_1(Symp(\mathbb{C} P^2 \# 5{\overline {\mathbb{C} P^2}}, ω)$ for any given symplectic form. Our results can be uniformly presented regarding Dynkin diagrams of type $\mathbb{A}$ and type $\mathbb{D}$ Lie algebras. We also provide a solution to the smooth isotopy problem of rational $4$-manifolds.

math.SG↗
arXiv2019

LPRNet: Lightweight Deep Network by Low-rank Pointwise Residual Convolution

Deep learning has become popular in recent years primarily due to the powerful computing device such as GPUs. However, deploying these deep models to end-user devices, smart phones, or embedded systems with limited resources is challenging. To reduce the computation and memory costs, we propose a novel lightweight deep learning module by low-rank pointwise residual (LPR) convolution, called LPRNet. Essentially, LPR aims at using low-rank approximation in pointwise convolution to further reduce the module size, while keeping depthwise convolutions as the residual module to rectify the LPR module. This is critical when the low-rankness undermines the convolution process. We embody our design by replacing modules of identical input-output dimension in MobileNet and ShuffleNetv2. Experiments on visual recognition tasks including image classification and face alignment on popular benchmarks show that our LPRNet achieves competitive performance but with significant reduction of Flops and memory cost compared to the state-of-the-art deep models focusing on model compression.

cs.CV↗
arXiv2019

Topological magnetic-spin textures in two-dimensional van der Waals Cr2Ge2Te6

Two-dimensional (2D) van der Waals (vdW) materials show a range of profound physical properties that can be tailored through their incorporation in heterostructures and manipulated with external forces. The recent discovery of long-range ferromagnetic order down to atomic layers provides an additional degree of freedom in engineering 2D materials and their heterostructure devices for spintronics, valleytronics and magnetic tunnel junction switches. Here, using direct imaging by cryo-Lorentz transmission electron microscopy we show that topologically nontrivial magnetic-spin states, skyrmionic bubbles, can be realized in exfoliated insulating 2D vdW Cr2Ge2Te6. Due to the competition between dipolar interactions and uniaxial magnetic anisotropy, hexagonally-packed nanoscale bubble lattices emerge by field cooling with magnetic field applied along the out-of-plane direction. Despite a range of topological spin textures in stripe domains arising due to pair formation and annihilation of Bloch lines, bubble lattices with single chirality are prevalent. Our observation of topologically-nontrivial homochiral skyrmionic bubbles in exfoliated vdW materials provides a new avenue for novel quantum states in atomically-thin insulators for magneto-electronic and quantum devices.

cond-mat.mtrl-sci↗
arXiv2019

Federated Learning with Differential Privacy: Algorithms and Performance Analysis

In this paper, to effectively prevent information leakage, we propose a novel framework based on the concept of differential privacy (DP), in which artificial noises are added to the parameters at the clients side before aggregating, namely, noising before model aggregation FL (NbAFL). First, we prove that the NbAFL can satisfy DP under distinct protection levels by properly adapting different variances of artificial noises. Then we develop a theoretical convergence bound of the loss function of the trained FL model in the NbAFL. Specifically, the theoretical bound reveals the following three key properties: 1) There is a tradeoff between the convergence performance and privacy protection levels, i.e., a better convergence performance leads to a lower protection level; 2) Given a fixed privacy protection level, increasing the number $N$ of overall clients participating in FL can improve the convergence performance; 3) There is an optimal number of maximum aggregation times (communication rounds) in terms of convergence performance for a given protection level. Furthermore, we propose a $K$-random scheduling strategy, where $K$ ($1<K<N$) clients are randomly selected from the $N$ overall clients to participate in each aggregation. We also develop the corresponding convergence bound of the loss function in this case and the $K$-random scheduling strategy can also retain the above three properties. Moreover, we find that there is an optimal $K$ that achieves the best convergence performance at a fixed privacy level. Evaluations demonstrate that our theoretical results are consistent with simulations, thereby facilitating the designs on various privacy-preserving FL algorithms with different tradeoff requirements on convergence performance and privacy levels.

cs.LG↗
arXiv2019

Privacy Preserving Location Data Publishing: A Machine Learning Approach

Publishing datasets plays an essential role in open data research and promoting transparency of government agencies. However, such data publication might reveal users' private information. One of the most sensitive sources of data is spatiotemporal trajectory datasets. Unfortunately, merely removing unique identifiers cannot preserve the privacy of users. Adversaries may know parts of the trajectories or be able to link the published dataset to other sources for the purpose of user identification. Therefore, it is crucial to apply privacy preserving techniques before the publication of spatiotemporal trajectory datasets. In this paper, we propose a robust framework for the anonymization of spatiotemporal trajectory datasets termed as machine learning based anonymization (MLA). By introducing a new formulation of the problem, we are able to apply machine learning algorithms for clustering the trajectories and propose to use $k$-means algorithm for this purpose. A variation of $k$-means algorithm is also proposed to preserve the privacy in overly sensitive datasets. Moreover, we improve the alignment process by considering multiple sequence alignment as part of the MLA. The framework and all the proposed algorithms are applied to TDrive and Geolife location datasets. The experimental results indicate a significantly higher utility of datasets by anonymization based on MLA framework.

cs.LG↗
arXiv2019

User Association and Path Planning for UAV-Aided Mobile Edge Computing with Energy Restriction

Mobile edge computing (MEC) provides computational services at the edge of networks by offloading tasks from user equipments (UEs). This letter employs an unmanned aerial vehicle (UAV) as the edge computing server to execute offloaded tasks from the ground UEs. We jointly optimize user association, UAV trajectory, and uploading power of each UE to maximize sum bits offloaded from all UEs to the UAV, subject to energy constraint of the UAV and quality of service (QoS) of each UE. To address the non-convex optimization problem, we first decompose it into three subproblems that are solved with integer programming and successive convex optimization methods respectively. Then, we tackle the overall problem by the multi-variable iterative optimization algorithm. Simulations show that the proposed algorithm can achieve a better performance than other baseline schemes.

cs.IT↗
arXiv2019

Bandwidth Slicing to Boost Federated Learning in Edge Computing

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.

cs.NI↗
arXiv2019

Dynamic Content Update for Wireless Edge Caching via Deep Reinforcement Learning

This letter studies a basic wireless caching network where a source server is connected to a cache-enabled base station (BS) that serves multiple requesting users. A critical problem is how to improve cache hit rate under dynamic content popularity. To solve this problem, the primary contribution of this work is to develop a novel dynamic content update strategy with the aid of deep reinforcement learning. Considering that the BS is unaware of content popularities, the proposed strategy dynamically updates the BS cache according to the time-varying requests and the BS cached contents. Towards this end, we model the problem of cache update as a Markov decision process and put forth an efficient algorithm that builds upon the long short-term memory network and external memory to enhance the decision making ability of the BS. Simulation results show that the proposed algorithm can achieve not only a higher average reward than deep Q-network, but also a higher cache hit rate than the existing replacement policies such as the least recently used, first-in first-out, and deep Q-network based algorithms.

cs.IT↗