arXiv subjects

Xinyuan Zhao

Publications and source records attributed to Xinyuan Zhao.

34 records · Page 2Linked to original sources

arXiv2023

Strong Variational Sufficiency for Nonlinear Semidefinite Programming and its Implications

Strong variational sufficiency is a newly proposed property, which turns out to be of great use in the convergence analysis of multiplier methods. However, what this property implies for non-polyhedral problems remains a puzzle. In this paper, we prove the equivalence between the strong variational sufficiency and the strong second order sufficient condition (SOSC) for nonlinear semidefinite programming (NLSDP), without requiring the uniqueness of multiplier or any other constraint qualifications. Based on this characterization, the local convergence property of the augmented Lagrangian method (ALM) for NLSDP can be established under strong SOSC in the absence of constraint qualifications. Moreover, under the strong SOSC, we can apply the semi-smooth Newton method to solve the ALM subproblems of NLSDP as the positive definiteness of the generalized Hessian of augmented Lagrangian function is satisfied.

math.OC↗

arXiv2023

A Hybrid Self-Supervised Learning Framework for Vertical Federated Learning

Vertical federated learning (VFL), a variant of Federated Learning (FL), has recently drawn increasing attention as the VFL matches the enterprises' demands of leveraging more valuable features to achieve better model performance. However, conventional VFL methods may run into data deficiency as they exploit only aligned and labeled samples (belonging to different parties), leaving often the majority of unaligned and unlabeled samples unused. The data deficiency hampers the effort of the federation. In this work, we propose a Federated Hybrid Self-Supervised Learning framework, named FedHSSL, that utilizes cross-party views (i.e., dispersed features) of samples aligned among parties and local views (i.e., augmentation) of unaligned samples within each party to improve the representation learning capability of the VFL joint model. FedHSSL further exploits invariant features across parties to boost the performance of the joint model through partial model aggregation. FedHSSL, as a framework, can work with various representative SSL methods. We empirically demonstrate that FedHSSL methods outperform baselines by large margins. We provide an in-depth analysis of FedHSSL regarding label leakage, which is rarely investigated in existing self-supervised VFL works. The experimental results show that, with proper protection, FedHSSL achieves the best privacy-utility trade-off against the state-of-the-art label inference attack compared with baselines. Code is available at \url{https://github.com/jorghyq2016/FedHSSL}.

cs.LG↗

arXiv2022

3DFill:Reference-guided Image Inpainting by Self-supervised 3D Image Alignment

Most existing image inpainting algorithms are based on a single view, struggling with large holes or the holes containing complicated scenes. Some reference-guided algorithms fill the hole by referring to another viewpoint image and use 2D image alignment. Due to the camera imaging process, simple 2D transformation is difficult to achieve a satisfactory result. In this paper, we propose 3DFill, a simple and efficient method for reference-guided image inpainting. Given a target image with arbitrary hole regions and a reference image from another viewpoint, the 3DFill first aligns the two images by a two-stage method: 3D projection + 2D transformation, which has better results than 2D image alignment. The 3D projection is an overall alignment between images and the 2D transformation is a local alignment focused on the hole region. The entire process of image alignment is self-supervised. We then fill the hole in the target image with the contents of the aligned image. Finally, we use a conditional generation network to refine the filled image to obtain the inpainting result. 3DFill achieves state-of-the-art performance on image inpainting across a variety of wide view shifts and has a faster inference speed than other inpainting models.

cs.CV↗

arXiv2022

Global Spectral Filter Memory Network for Video Object Segmentation

This paper studies semi-supervised video object segmentation through boosting intra-frame interaction. Recent memory network-based methods focus on exploiting inter-frame temporal reference while paying little attention to intra-frame spatial dependency. Specifically, these segmentation model tends to be susceptible to interference from unrelated nontarget objects in a certain frame. To this end, we propose Global Spectral Filter Memory network (GSFM), which improves intra-frame interaction through learning long-term spatial dependencies in the spectral domain. The key components of GSFM is 2D (inverse) discrete Fourier transform for spatial information mixing. Besides, we empirically find low frequency feature should be enhanced in encoder (backbone) while high frequency for decoder (segmentation head). We attribute this to semantic information extracting role for encoder and fine-grained details highlighting role for decoder. Thus, Low (High) Frequency Module is proposed to fit this circumstance. Extensive experiments on the popular DAVIS and YouTube-VOS benchmarks demonstrate that GSFM noticeably outperforms the baseline method and achieves state-of-the-art performance. Besides, extensive analysis shows that the proposed modules are reasonable and of great generalization ability. Our source code is available at https://github.com/workforai/GSFM.

cs.CV↗

arXiv2022

Learning Quality-aware Dynamic Memory for Video Object Segmentation

Recently, several spatial-temporal memory-based methods have verified that storing intermediate frames and their masks as memory are helpful to segment target objects in videos. However, they mainly focus on better matching between the current frame and the memory frames without explicitly paying attention to the quality of the memory. Therefore, frames with poor segmentation masks are prone to be memorized, which leads to a segmentation mask error accumulation problem and further affect the segmentation performance. In addition, the linear increase of memory frames with the growth of frame number also limits the ability of the models to handle long videos. To this end, we propose a Quality-aware Dynamic Memory Network (QDMN) to evaluate the segmentation quality of each frame, allowing the memory bank to selectively store accurately segmented frames to prevent the error accumulation problem. Then, we combine the segmentation quality with temporal consistency to dynamically update the memory bank to improve the practicability of the models. Without any bells and whistles, our QDMN achieves new state-of-the-art performance on both DAVIS and YouTube-VOS benchmarks. Moreover, extensive experiments demonstrate that the proposed Quality Assessment Module (QAM) can be applied to memory-based methods as generic plugins and significantly improves performance. Our source code is available at https://github.com/workforai/QDMN.

cs.CV↗

arXiv2021

A Unified Convergence Analysis of a Second-Order Method of Multipliers for Nonlinear Conic Programming

In this paper, we accomplish a unified convergence analysis of a second-order method of multipliers (i.e., a second-order augmented Lagrangian method) for solving the conventional nonlinear conic optimization problems.Specifically, the algorithm that we investigated incorporates a specially designed nonsmooth (generalized) Newton step to furnish a second-order update rule for the multipliers.We first show in a unified fashion that under a few abstract assumptions, the proposed method is locally convergent and possesses a (nonasymptotic) superlinear convergence rate, even though the penalty parameter is fixed and/or the strict complementarity fails.Subsequently, we demonstrate that, for the three typical scenarios, i.e., the classic nonlinear programming, the nonlinear second-order cone programming, and the nonlinear semidefinite programming, these abstract assumptions are nothing but exactly the implications of the iconic sufficient conditions that were assumed for establishing the Q-linear convergence rates of the method of multipliers without assuming the strict complementarity.

math.OC↗

arXiv2021

Real-time Human-Centric Segmentation for Complex Video Scenes

Most existing video tasks related to "human" focus on the segmentation of salient humans, ignoring the unspecified others in the video. Few studies have focused on segmenting and tracking all humans in a complex video, including pedestrians and humans of other states (e.g., seated, riding, or occluded). In this paper, we propose a novel framework, abbreviated as HVISNet, that segments and tracks all presented people in given videos based on a one-stage detector. To better evaluate complex scenes, we offer a new benchmark called HVIS (Human Video Instance Segmentation), which comprises 1447 human instance masks in 805 high-resolution videos in diverse scenes. Extensive experiments show that our proposed HVISNet outperforms the state-of-the-art methods in terms of accuracy at a real-time inference speed (30 FPS), especially on complex video scenes. We also notice that using the center of the bounding box to distinguish different individuals severely deteriorates the segmentation accuracy, especially in heavily occluded conditions. This common phenomenon is referred to as the ambiguous positive samples problem. To alleviate this problem, we propose a mechanism named Inner Center Sampling to improve the accuracy of instance segmentation. Such a plug-and-play inner center sampling mechanism can be incorporated in any instance segmentation models based on a one-stage detector to improve the performance. In particular, it gains 4.1 mAP improvement on the state-of-the-art method in the case of occluded humans. Code and data are available at https://github.com/IIGROUP/HVISNet.

cs.CV↗

arXiv2021

PoseDet: Fast Multi-Person Pose Estimation Using Pose Embedding

Current methods of multi-person pose estimation typically treat the localization and the association of body joints separately. It is convenient but inefficient, leading to additional computation and a waste of time. This paper, however, presents a novel framework PoseDet (Estimating Pose by Detection) to localize and associate body joints simultaneously at higher inference speed. Moreover, we propose the keypoint-aware pose embedding to represent an object in terms of the locations of its keypoints. The proposed pose embedding contains semantic and geometric information, allowing us to access discriminative and informative features efficiently. It is utilized for candidate classification and body joint localization in PoseDet, leading to robust predictions of various poses. This simple framework achieves an unprecedented speed and a competitive accuracy on the COCO benchmark compared with state-of-the-art methods. Extensive experiments on the CrowdPose benchmark show the robustness in the crowd scenes. Source code is available.

cs.CV↗

arXiv2020

Matrix optimization based Euclidean embedding with outliers

Euclidean embedding from noisy observations containing outlier errors is an important and challenging problem in statistics and machine learning. Many existing methods would struggle with outliers due to a lack of detection ability. In this paper, we propose a matrix optimization based embedding model that can produce reliable embeddings and identify the outliers jointly. We show that the estimators obtained by the proposed method satisfy a non-asymptotic risk bound, implying that the model provides a high accuracy estimator with high probability when the order of the sample size is roughly the degree of freedom up to a logarithmic factor. Moreover, we show that under some mild conditions, the proposed model also can identify the outliers without any prior information with high probability. Finally, numerical experiments demonstrate that the matrix optimization-based model can produce configurations of high quality and successfully identify outliers even for large networks.

stat.ML↗

arXiv2019

A proximal DC approach for quadratic assignment problem

In this paper, we show that the quadratic assignment problem (QAP) can be reformulated to an equivalent rank constrained doubly nonnegative (DNN) problem. Under the framework of the difference of convex functions (DC) approach, a semi-proximal DC algorithm (DCA) is proposed for solving the relaxation of the rank constrained DNN problem whose subproblems can be solved by the semi-proximal augmented Lagrangian method (sPALM). We show that the generated sequence converges to a stationary point of the corresponding DC problem, which is feasible to the rank constrained DNN problem. Moreover, numerical experiments demonstrate that for most QAP instances, the proposed approach can find the global optimal solutions efficiently, and for others, the proposed algorithm is able to provide good feasible solutions in a reasonable time.

math.OC↗

arXiv2017

Quadratic growth conditions for convex matrix optimization problems associated with spectral functions

In this paper, we provide two types of sufficient conditions for ensuring the quadratic growth conditions of a class of constrained convex symmetric and non-symmetric matrix optimization problems regularized by nonsmooth spectral functions. These sufficient conditions are derived via the study of the $\mathcal{C}^2$-cone reducibility of spectral functions and the metric subregularity of their subdifferentials, respectively. As an application, we demonstrate how quadratic growth conditions are used to guarantee the desirable fast convergence rates of the augmented Lagrangian methods (ALM) for solving convex matrix optimization problems. Numerical experiments on an easy-to-implement ALM applied to the fastest mixing Markov chain problem are also presented to illustrate the significance of the obtained results.

math.OC↗

arXiv2004

Amorphous ZrO2 from Ab-initio molecular dynamics: Structural, electronic and dielectric properties

Realistic models of amorphous ZrO2 are generated in a ``melt-and-quench'' fashion using ab-initio molecular dynamics in a plane-wave pseudopotential formulation of density-functional theory. The structural properties of the resulting amorphous models are analyzed, with special attention to coordination statistics. The vibrational and dielectric properties of one of these models are then investigated from first principles using linear-response methods. The electronic dielectric constant and Born effective charges are found to be very similar to those of the crystalline phases. Encouragingly, the predicted total static dielectric constant is about 22, comparable to that of the monoclinic phase. This work is motivated by the search for improved gate dielectric materials for sub-0.1 micron CMOS technology, and may also have implications for HfO2 and for silicates of ZrO2 and HfO2.

cond-mat.mtrl-sci↗

arXiv2003

First-principles Study of Electronic and Dielectric Properties of ZrO2 and HfO2

Using density-functional theory with ultrasoft pseudopotentials, we previously investigated the structural and electronic properties of the low-pressure (cubic, tetragonal, and monoclinic) phases of ZrO2 and HfO2, in order to elucidate phonon modes, Born effective charge tensors, and especially the lattice dielectric response in these phases. We now extend this previous work by carrying out similar calculations on the two high-pressure orthorhombic phases, and by providing density-of-states and band-gap information on all polymorphs. Our results show that the electronic structures and dielectric responses are strongly phase-dependent. In particular, the monoclinic phases of ZrO2 and HfO2 are found to have a strongly anisotropic dielectric tensor and a rather small orientational average (epsilon_0) compared to the two other low-pressure phases. Our calculations show that epsilon_0 is even smaller in the orthorhombic phases.

cond-mat.mtrl-sci↗

arXiv2002

First-principles study of structural, vibrational and lattice dielectric properties of hafnium oxide

Crystalline structures, zone-center phonon modes, and the related dielectric response of the three low-pressure phases of HfO2 have been investigated in density-functional theory using ultrasoft pseudopotentials and a plane-wave basis. The structures of low-pressure HfO2 polymorphs are carefully studied with both the local-density approximation (LDA) and the generalized gradient approximation (GGA). The fully relaxed structures obtained with either exchange-correlation scheme agree reasonably well with experiment, although LDA yields better overall agreement. After calculating the Born effective charge tensors and the force-constant matrices by finite-difference methods, the lattice dielectric susceptibility tensors for the three HfO2 phases are computed by decomposing the tensors into the contributions from individual infrared-active phonon modes.

cond-mat.mtrl-sci↗

arXiv2001

Phonons and Lattice Dielectric Properties of Zirconia

We have performed a first-principles study of the structural and vibrational properties of the three low-pressure (cubic, tetragonal, and especially monoclinic) phases of ZrO2, with special attention to the computation of the zone-center phonon modes and related dielectric properties. The calculations have been carried out within the local-density approximation using ultrasoft pseudopotentials and a plane-wave basis. The fully relaxed structural parameters are found to be in excellent agreement with experimental data and with previous theoretical work. The total-energy calculations correctly reproduce the energetics of the ZrO2 phases, and the calculated zone-center phonon frequencies yield good agreement with the infrared and Raman experimental frequencies in the monoclinic phase. The Born effective charge tensors are computed and, together with the mode eigenvectors, used to decompose the lattice dielectric susceptibility tensor into contributions arising from individual infrared-active phonon modes. This work has been partially motivated by the potential for ZrO2 to replace SiO2 as the gate-dielectric material in modern integrated-circuit technology.

cond-mat.mtrl-sci↗

arXiv2000

Wannier-function description of the electronic polarization and infrared absorption of high-pressure hydrogen

We have constructed maximally-localized Wannier functions for prototype structures of solid molecular hydrogen under pressure, starting from LDA and tight-binding Bloch wave functions. Each occupied Wannier function can be associated with two paired protons, defining a ``Wannier molecule''. The sum of the dipole moments of these ``molecules'' always gives the correct macroscopic polarization, even under strong compression, when the overlap between nearby Wannier functions becomes significant. We find that at megabar pressures the contributions to the dipoles arising from the overlapping tails of the Wannier functions is very large. The strong vibron infrared absorption experimentally observed in phase III, above ~ 150 GPa, is analyzed in terms of the vibron-induced fluctuations of the Wannier dipoles. We decompose these fluctuations into ``static'' and ``dynamical'' contributions, and find that at such high densities the latter term, which increases much more steeply with pressure, is dominant.

cond-mat.mtrl-sci↗