arXiv subjects

Zheng Zhu

Publications and source records attributed to Zheng Zhu.

At least 271 records · Page 15Linked to original sources

arXiv2018

Distractor-aware Siamese Networks for Visual Object Tracking

Recently, Siamese networks have drawn great attention in visual tracking community because of their balanced accuracy and speed. However, features used in most Siamese tracking approaches can only discriminate foreground from the non-semantic backgrounds. The semantic backgrounds are always considered as distractors, which hinders the robustness of Siamese trackers. In this paper, we focus on learning distractor-aware Siamese networks for accurate and long-term tracking. To this end, features used in traditional Siamese trackers are analyzed at first. We observe that the imbalanced distribution of training data makes the learned features less discriminative. During the off-line training phase, an effective sampling strategy is introduced to control this distribution and make the model focus on the semantic distractors. During inference, a novel distractor-aware module is designed to perform incremental learning, which can effectively transfer the general embedding to the current video domain. In addition, we extend the proposed approach for long-term tracking by introducing a simple yet effective local-to-global search region strategy. Extensive experiments on benchmarks show that our approach significantly outperforms the state-of-the-arts, yielding 9.6% relative gain in VOT2016 dataset and 35.9% relative gain in UAV20L dataset. The proposed tracker can perform at 160 FPS on short-term benchmarks and 110 FPS on long-term benchmarks.

cs.CV↗

arXiv2018

Multichannel charge Kondo effect and non-Fermi liquid fixed points in conventional and topological superconductor islands

We study multiterminal Majorana and conventional superconducting islands in the vicinity of the charge degeneracy point using bosonization and numerical renormalization group. Both models map to the multichannel charge Kondo problem, but for noninteracting normal leads they flow to different non-Fermi liquid fixed points at low temperatures. We compare and contrast both cases by numerically obtaining the full crossover to the low temperature regime and predict distinctive transport signatures. We attribute the differences between both types of islands to a crucial distinction of charge-$2e$ and charge-$e$ transfer in the conventional and topological case, respectively. In the conventional case, our results establish s-wave islands as a new platform to study the intermediate multichannel Kondo fixed point. In the topological setup the crossover temperature to non-Fermi liquid behavior is relatively high as it is proportional to level broadening and the transport results are not sensitive to channel coupling anisotropy, moving away from the charge degeneracy point or including a small Majorana hybridization, which makes our proposal experimentally feasible.

cond-mat.mes-hall↗

arXiv2018

Optical Flow Based Real-time Moving Object Detection in Unconstrained Scenes

Real-time moving object detection in unconstrained scenes is a difficult task due to dynamic background, changing foreground appearance and limited computational resource. In this paper, an optical flow based moving object detection framework is proposed to address this problem. We utilize homography matrixes to online construct a background model in the form of optical flow. When judging out moving foregrounds from scenes, a dual-mode judge mechanism is designed to heighten the system's adaptation to challenging situations. In experiment part, two evaluation metrics are redefined for more properly reflecting the performance of methods. We quantitatively and qualitatively validate the effectiveness and feasibility of our method with videos in various scene conditions. The experimental results show that our method adapts itself to different situations and outperforms the state-of-the-art methods, indicating the advantages of optical flow based methods.

cs.CV↗

arXiv2018

Evaluation of adaptive treatment strategies in an observational study where time-varying covariates are not monitored systematically

In studies based on electronic health records (EHR), the frequency of covariate monitoring can vary by covariate type, across patients, and over time. This can lead to major challenges: first, the difference in monitoring protocols may invalidate the extrapolation of study results obtained in one population to the other, and second, monitoring can act as a time-varying confounder of the causal effect of a time-varying treatment on the outcomes of interest. This paper demonstrates how to account for non-systematic covariate monitoring when evaluating dynamic treatment interventions, and how to evaluate joint dynamic treatment-censoring and static monitoring interventions, in a real world, EHR-based, comparative effectiveness research (CER) study of patients with type II diabetes mellitus. First, we show that the effects of dynamic treatment-censoring regimes can be identified by including indicators of monitoring events in the adjustment set. Second, we demonstrate the poor performance of the standard inverse probability weighting (IPW) estimator of the effects of joint treatment-censoring-monitoring interventions, due to a large decrease in data support resulting in a large increase in standard errors and concerns over finite-sample bias from near-violations of the positivity assumption for the monitoring process. Finally, we detail an alternate IPW estimator of the effects of these interventions using the No Direct Effect assumption. We demonstrate that this estimator can result in improved efficiency but at the cost of increased bias concerns over structural near-violations of the positivity assumption for the treatment process. To conclude, this paper develops and illustrates new tools that researchers can exploit to appropriately account for non-systematic covariate monitoring in CER, and to ask new causal questions about the joint effects of treatment and monitoring interventions.

stat.ME↗

arXiv2018

Hidden spin current in doped Mott antiferromagnets

We investigate the nature of doped Mott insulators using exact diagonalization and density matrix renormalization group methods. Persistent spin currents are revealed in the ground state, which are concomitant with a nonzero total momentum or angular momentum associated with the doped hole. The latter determines a nontrivial ground state degeneracy. By further making superpositions of the degenerate ground states with zero or unidirectional spin currents, we show that different patterns of spatial charge and spin modulations will emerge. Such anomaly persists for the odd numbers of holes, but the spin current, ground state degeneracy, and charge/spin modulations completely disappear for even numbers of holes, with the two-hole ground state exhibiting a d-wave symmetry. An understanding of the spin current due to a many-body Berry-like phase and its impact on the momentum distribution of the doped holes will be discussed.

cond-mat.str-el↗

arXiv2018

Robust non-Abelian spin liquid and possible intermediate phase in antiferromagnetic Kitaev model with magnetic field

We investigate the non-Abelian topological chiral spin liquid phase in the two-dimensional (2D) Kitaev honeycomb model subject to a magnetic field. By combining density matrix renormalization group (DMRG) and exact diagonalization (ED) we study the energy spectra, entanglement, topological degeneracy, and expectation values of Wilson loop operators, allowing for robust characterization. While the ferromagnetic (FM) Kitaev spin liquid is already destroyed by a weak magnetic field with Zeeman energy $H_*^\text{FM} \approx 0.02$, the antiferromagnetic (AFM) spin liquid remains robust up to a magnetic field that is an order of magnitude larger, $H_*^\text{AFM} \approx 0.2$. Interestingly, for larger fields $H_*^\text{AFM} < H < H_{**}^\text{AFM}$, an intermediate gapless phase is observed, before a second transition to the high-field partially-polarized paramagnet. We attribute this rich phase diagram, and the remarkable stability of the chiral topological phase in the AFM Kitaev model, to the interplay of strong spin-orbit coupling and frustration enhanced by the magnetic field. Our findings suggest relevance to recent experiments on RuCl$_3$ under magnetic fields.

cond-mat.str-el↗

arXiv2018

Lack of a thermodynamic finite-temperature spin-glass phase in the two-dimensional randomly-coupled ferromagnet

The search for problems where quantum adiabatic optimization might excel over classical optimization techniques has sparked a recent interest in inducing a finite-temperature spin-glass transition in quasi-planar topologies. We have performed large-scale finite-temperature Monte Carlo simulations of a two-dimensional square-lattice bimodal spin glass with next-nearest ferromagnetic interactions claimed to exhibit a finite-temperature spin-glass state for a particular relative strength of the next-nearest to nearest interactions [Phys. Rev. Lett. 76, 4616 (1996)]. Our results show that the system is in a paramagnetic state in the thermodynamic limit, despite zero-temperature simulations [Phys. Rev. B 63, 094423 (2001)] suggesting the existence of a finite-temperature spin-glass transition. Therefore, deducing the finite-temperature behavior from zero-temperature simulations can be dangerous when corrections to scaling are large.

cond-mat.dis-nn↗

arXiv2018

End-to-end Video-level Representation Learning for Action Recognition

From the frame/clip-level feature learning to the video-level representation building, deep learning methods in action recognition have developed rapidly in recent years. However, current methods suffer from the confusion caused by partial observation training, or without end-to-end learning, or restricted to single temporal scale modeling and so on. In this paper, we build upon two-stream ConvNets and propose Deep networks with Temporal Pyramid Pooling (DTPP), an end-to-end video-level representation learning approach, to address these problems. Specifically, at first, RGB images and optical flow stacks are sparsely sampled across the whole video. Then a temporal pyramid pooling layer is used to aggregate the frame-level features which consist of spatial and temporal cues. Lastly, the trained model has compact video-level representation with multiple temporal scales, which is both global and sequence-aware. Experimental results show that DTPP achieves the state-of-the-art performance on two challenging video action datasets: UCF101 and HMDB51, either by ImageNet pre-training or Kinetics pre-training.

cs.CV↗

arXiv2018

Pairing versus phase coherence of doped holes in distinct quantum spin backgrounds

We examine the pairing structure of holes injected into two \emph{distinct} spin backgrounds: a short-range antiferromagnetic phase versus a symmetry protected topological phase. Based on density matrix renormalization group (DMRG) simulation, we find that although there is a strong binding between two holes in both phases, \emph{phase fluctuations} can significantly influence the pair-pair correlation depending on the spin-spin correlation in the background. Here the phase fluctuation is identified as an intrinsic string operator nonlocally controlled by the spins. We show that while the pairing amplitude is generally large, the coherent Cooper pairing can be substantially weakened by the phase fluctuation in the symmetry-protected topological phase, in contrast to the short-range antiferromagnetic phase. It provides an example of a non-BCS mechanism for pairing, in which the paring phase coherence is determined by the underlying spin state self-consistently, bearing an interesting resemblance to the pseudogap physics in the cuprate.

cond-mat.str-el↗

arXiv2018

End-to-end Flow Correlation Tracking with Spatial-temporal Attention

Discriminative correlation filters (DCF) with deep convolutional features have achieved favorable performance in recent tracking benchmarks. However, most of existing DCF trackers only consider appearance features of current frame, and hardly benefit from motion and inter-frame information. The lack of temporal information degrades the tracking performance during challenges such as partial occlusion and deformation. In this work, we focus on making use of the rich flow information in consecutive frames to improve the feature representation and the tracking accuracy. Firstly, individual components, including optical flow estimation, feature extraction, aggregation and correlation filter tracking are formulated as special layers in network. To the best of our knowledge, this is the first work to jointly train flow and tracking task in a deep learning framework. Then the historical feature maps at predefined intervals are warped and aggregated with current ones by the guiding of flow. For adaptive aggregation, we propose a novel spatial-temporal attention mechanism. Extensive experiments are performed on four challenging tracking datasets: OTB2013, OTB2015, VOT2015 and VOT2016, and the proposed method achieves superior results on these benchmarks.

cs.CV↗

arXiv2018

NAE-SAT-based probabilistic membership filters

Probabilistic membership filters are a type of data structure designed to quickly verify whether an element of a large data set belongs to a subset of the data. While false negatives are not possible, false positives are. Therefore, the main goal of any good probabilistic membership filter is to have a small false-positive rate while being memory efficient and fast to query. Although Bloom filters are fast to construct, their memory efficiency is bounded by a strict theoretical upper bound. Weaver et al. introduced random satisfiability-based filters that significantly improved the efficiency of the probabilistic filters, however, at the cost of solving a complex random satisfiability (SAT) formula when constructing the filter. Here we present an improved SAT filter approach with a focus on reducing the filter building times, as well as query times. Our approach is based on using not-all-equal (NAE) SAT formulas to build the filters, solving these via a mapping to random SAT using traditionally-fast random SAT solvers, as well as bit packing and the reduction of the number of hash functions. Paired with fast hardware, NAE-SAT filters could result in enterprise-size applications.

cs.DS↗

arXiv2017

Quantum Hall ferroelectrics and nematics in multivalley systems

We study broken symmetry states at integer Landau level fillings in multivalley quantum Hall systems whose low energy dispersions are anisotropic. When the Fermi surface of individual pockets lacks twofold rotational symmetry, like in Bismuth (111) and in Sn$_{1-x}$Pb$_x$Se (001) surfaces, interactions tend to drive the formation of quantum Hall ferroelectric states. We demonstrate that the dipole moment in these states has an intimate relation to the Fermi surface geometry of the parent metal. In quantum Hall nematic states, like those arising in AlAs quantum wells, we demonstrate the existence of unusually robust skyrmion quasiparticles.

cond-mat.str-el↗

arXiv2017

UCT: Learning Unified Convolutional Networks for Real-time Visual Tracking

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.

cs.CV↗

arXiv2017

Anisotropy Driven Transition from Moore-Read State to Quantum Hall Stripes

We investigate the nature of the quantum Hall liquid in a half-filled second Landau level ($n=1$) as a function of band mass anisotropy using numerical exact diagonalization (ED) and density matrix renormalization group (DMRG) methods. We find increasing the mass anisotropy induces a quantum phase transition from the Moore-Read state to a charge density wave state. By analyzing the energy spectrum, guiding center structure factors and by adding weak pinning potentials, we show that this charge density wave is a uni-directional quantum Hall stripe, which has a periodicity of a few magnetic lengths and survives in the thermodynamic limit. We find smooth profiles for the guiding center occupation function that reveal the strong coupling nature of the array of chiral Luttinger liquids residing at the stripe edges.

cond-mat.str-el↗

arXiv2017

Numerical Study of Quantum Hall Bilayers at Total Filling $ν_T=1$: A New Phase at Intermediate Layer Distances

We study the phase diagram of quantum Hall bilayer systems with total filing $ν_T=1/2+1/2$ of the lowest Landau level as a function of layer distances $d$. Based on numerical exact diagonalization calculations, we obtain three distinct phases, including an exciton superfluid phase with spontaneous interlayer coherence at small $d$, a composite Fermi liquid at large $d$, and an intermediate phase for $1.1<d/l_B<1.8$ ($l_B$ is the magnetic length). The transition from the exciton superfluid to the intermediate phase is identified by (i) a dramatic change in the Berry curvature of the ground state under twisted boundary conditions on the two layers; (ii) an energy level crossing of the first excited state. The transition from the intermediate phase to the composite Fermi liquid is identified by the vanishing of the exciton superfluid stiffness. Furthermore, from our finite-size study, the energy cost of transferring one electron between the layers shows an even-odd effect and possibly extrapolates to a finite value in the thermodynamic limit, indicating the enhanced intralayer correlation. Our identification of an intermediate phase and its distinctive features shed new light on the theoretical understanding of the quantum Hall bilayer system at total filling $ν_T=1$.

cond-mat.str-el↗

arXiv2017

Learning Gating ConvNet for Two-Stream based Methods in Action Recognition

For the two-stream style methods in action recognition, fusing the two streams' predictions is always by the weighted averaging scheme. This fusion method with fixed weights lacks of pertinence to different action videos and always needs trial and error on the validation set. In order to enhance the adaptability of two-stream ConvNets and improve its performance, an end-to-end trainable gated fusion method, namely gating ConvNet, for the two-stream ConvNets is proposed in this paper based on the MoE (Mixture of Experts) theory. The gating ConvNet takes the combination of feature maps from the same layer of the spatial and the temporal nets as input and adopts ReLU (Rectified Linear Unit) as the gating output activation function. To reduce the over-fitting of gating ConvNet caused by the redundancy of parameters, a new multi-task learning method is designed, which jointly learns the gating fusion weights for the two streams and learns the gating ConvNet for action classification. With our gated fusion method and multi-task learning approach, a high accuracy of 94.5% is achieved on the dataset UCF101.

cs.CV↗

arXiv2017

Universal properties of the FQH state from the topological entanglement entropy and disorder effects

The topological entanglement entropy (TEE) is a robust measurement of the quantum many-body state with topological order. In fractional quantum Hall (FQH) state, it has a connection to the quantum dimension of the state itself and its quasihole excitations from the conformal field theory (CFT) description. We study the entanglement entropy (EE) in the Moore-Read (MR) and Read-Rezayi (RR) FQH states. The non-Abelian quasi- hole excitation induces an extra correction of the TEE which is related to its quantum dimension. With considering the effects of the disorder, the ground state TEE is stable before the spectral gap closing and the level statistics seems to have significant change with a stronger disorder, which indicates a many-body localization (MBL) transition.

cond-mat.str-el↗

arXiv2017

Exponentially-Biased Ground-State Sampling of Quantum Annealing Machines with Transverse-Field Driving Hamiltonians

We study the performance of the D-Wave 2X quantum annealing machine on systems with well-controlled ground-state degeneracy. While obtaining the ground state of a spin-glass benchmark instance represents a difficult task, the gold standard for any optimization algorithm or machine is to sample all solutions that minimize the Hamiltonian with more or less equal probability. Our results show that while naive transverse-field quantum annealing on the D-Wave 2X device can find the ground-state energy of the problems, it is not well suited in identifying all degenerate ground-state configurations associated to a particular instance. Even worse, some states are exponentially suppressed, in agreement with previous studies on toy model problems [New J. Phys. 11, 073021 (2009)]. These results suggest that more complex driving Hamiltonians are needed in future quantum annealing machines to ensure a fair sampling of the ground-state manifold.

quant-ph↗