Trust snapshot

Quick read

Trust 21 - EmergingVerification L1Unclaimed author
21works
0followers
18topics
4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Identity and collaboration

How to connect with this researcher

Claiming links this public author record to a researcher profile and unlocks direct collaboration workflows.

Log in to claim

Direct collaboration

Open a focused conversation when the fit is right

Claim this author entity first to unlock direct invitations.

Research graph

See the researcher in context

Open full explorer

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

21 published item(s)

preprint2026arXiv

Approximation Error Upper and Lower Bounds for Hölder Class with Transformers

We explore the expressive power of Transformers by establishing precise approximation error upper and lower bounds for Hölder class. Specifically, a new approximation upper bound is derived for the standard Transformer architecture equipped with Softmax operators, ReLU activation functions, and residual connections. We prove that a Transformer network composed of at most $\mathcal{O}(\varepsilon^{-{d_{0}}/α})$ blocks can approximate any bounded Hölder function with $d_{0}$-dimensional input and smoothness $α\in(0,1]$ under any accuracy $\varepsilon>0$. In the case of approximation lower bounds, leveraging the VC-dimension upper bound, we are the first to rigorously prove that Transformers demand for at least $Ω(\varepsilon^{-{d_{0}}/({4α})})$ blocks to achieve the $\varepsilon$ approximation accuracy. As a final step, we extend the derived results for standard Transformers to a general regression task and establish the corresponding excess risk rates demonstrating Transformers' empirical effectiveness in real-world settings.

preprint2022arXiv

"Second-Order Primal'' + "First-Order Dual'' Dynamical Systems with Time Scaling for Linear Equality Constrained Convex Optimization Problems

Second-order dynamical systems are important tools for solving optimization problems, and most of existing works in this field have focused on unconstrained optimization problems. In this paper, we propose an inertial primal-dual dynamical system with constant viscous damping and time scaling for the linear equality constrained convex optimization problem, which consists of a second-order ODE for the primal variable and a first-order ODE for the dual variable. When the scaling satisfies certain conditions, we prove its convergence property without assuming strong convexity. Even the convergence rate can become exponential when the scaling grows exponentially. We also show that the obtained convergence property of the dynamical system is preserved under a small perturbation.

preprint2022arXiv

EAGAN: Efficient Two-stage Evolutionary Architecture Search for GANs

Generative adversarial networks (GANs) have proven successful in image generation tasks. However, GAN training is inherently unstable. Although many works try to stabilize it by manually modifying GAN architecture, it requires much expertise. Neural architecture search (NAS) has become an attractive solution to search GANs automatically. The early NAS-GANs search only generators to reduce search complexity but lead to a sub-optimal GAN. Some recent works try to search both generator (G) and discriminator (D), but they suffer from the instability of GAN training. To alleviate the instability, we propose an efficient two-stage evolutionary algorithm-based NAS framework to search GANs, namely EAGAN. We decouple the search of G and D into two stages, where stage-1 searches G with a fixed D and adopts the many-to-one training strategy, and stage-2 searches D with the optimal G found in stage-1 and adopts the one-to-one training and weight-resetting strategies to enhance the stability of GAN training. Both stages use the non-dominated sorting method to produce Pareto-front architectures under multiple objectives (e.g., model size, Inception Score (IS), and Fréchet Inception Distance (FID)). EAGAN is applied to the unconditional image generation task and can efficiently finish the search on the CIFAR-10 dataset in 1.2 GPU days. Our searched GANs achieve competitive results (IS=8.81$\pm$0.10, FID=9.91) on the CIFAR-10 dataset and surpass prior NAS-GANs on the STL-10 dataset (IS=10.44$\pm$0.087, FID=22.18). Source code: https://github.com/marsggbo/EAGAN.

preprint2022arXiv

Evolutionary Multi-objective Architecture Search Framework: Application to COVID-19 3D CT Classification

The COVID-19 pandemic has threatened global health. Many studies have applied deep convolutional neural networks (CNN) to recognize COVID-19 based on chest 3D computed tomography (CT). Recent works show that no model generalizes well across CT datasets from different countries, and manually designing models for specific datasets requires expertise; thus, neural architecture search (NAS) that aims to search models automatically has become an attractive solution. To reduce the search cost on large 3D CT datasets, most NAS-based works use the weight-sharing (WS) strategy to make all models share weights within a supernet; however, WS inevitably incurs search instability, leading to inaccurate model estimation. In this work, we propose an efficient Evolutionary Multi-objective ARchitecture Search (EMARS) framework. We propose a new objective, namely potential, which can help exploit promising models to indirectly reduce the number of models involved in weights training, thus alleviating search instability. We demonstrate that under objectives of accuracy and potential, EMARS can balance exploitation and exploration, i.e., reducing search time and finding better models. Our searched models are small and perform better than prior works on three public COVID-19 3D CT datasets.

preprint2022arXiv

Fast primal-dual algorithm via dynamical system for a linearly constrained convex optimization problem

By time discretization of a second-order primal-dual dynamical system with damping $α/t$ where an inertial construction in the sense of Nesterov is needed only for the primal variable, we propose a fast primal-dual algorithm for a linear equality constrained convex optimization problem. Under a suitable scaling condition, we show that the proposed algorithm enjoys a fast convergence rate for the objective residual and the feasibility violation, and the decay rate can reach $\mathcal{O}(1/k^{α-1})$ at the most. We also study convergence properties of the corresponding primal-dual dynamical system to better understand the acceleration scheme. Finally, we report numerical experiments to demonstrate the effectiveness of the proposed algorithm.

preprint2022arXiv

Inorganic Crystal Structure Prototype Database based on Unsupervised Learning of Local Atomic Environments

Recognition of structure prototypes from tremendous known inorganic crystal structures has been an important subject beneficial for material science research and new materials design. The existing databases of inorganic crystal structure prototypes were mostly constructed by classifying materials in terms of the crystallographic space group information. Herein, we employed a distinct strategy to construct the inorganic crystal structure prototype database, relying on the classification of materials in terms of local atomic environments (LAE) accompanied by unsupervised machine learning method. Specifically, we adopted a hierarchical clustering approach onto all experimentally known inorganic crystal structures data to identify structure prototypes. The criterion for hierarchical clustering is the LAE represented by the state-of-the-art structure fingerprints of the improved bond-orientational order parameters and the smooth overlap of atomic positions. This allows us to build up a LAE-based Inorganic Crystal Structure Prototype Database (LAE-ICSPD) containing 15,613 structure prototypes with defined stoichiometries. In addition, we have developed a Structure Prototype Generator Infrastructure (SPGI) package, which is a useful toolkit for structure prototype generation. Our developed SPGI toolkit and LAE-ICSPD are beneficial for investigating inorganic materials in a global way as well as accelerating materials discovery process in the data-driven mode.

preprint2022arXiv

Negative Zero-Point-Energy Parameter in the Meyer-Miller Mapping Model for Nonadiabatic Dynamics

The celebrated Meyer-Miller mapping model has been a useful approach for generating practical trajectory-based nonadiabatic dynamics methods. It is generally assumed that the zero-point-energy (ZPE) parameter is positive. The constraint implied in the conventional Meyer-Miller mapping Hamiltonian for an F-electronic-state system actually requires that parameter γis larger than -1/F for the ZPE parameter for each electronic degree of freedom. Both negative and positive values are possible for such a parameter. We first establish a rigorous formulation to construct exact mapping models in the Cartesian phase space when the constraint is applied. When nuclear dynamics is approximated by the linearized semiclassical initial value representation, a negative ZPE parameter could lead to reasonably good performance in describing dynamic behaviors in typical spin-boson models for condensed-phase two-state systems, even at challenging zero temperature.

preprint2022arXiv

New Phase Space Formulations and Quantum Dynamics Approaches

We report recent progress on the phase space formulation of quantum mechanics with coordinate-momentum variables, focusing more on new theory of (weighted) constraint coordinate-momentum phase space for discrete-variable quantum systems. This leads to a general coordinate-momentum phase space formulation of composite quantum systems, where conventional representations on infinite phase space are employed for continuous variables. It is convenient to utilize (weighted) constraint coordinate-momentum phase space for representing the quantum state and describing nonclassical features. Various numerical tests demonstrate that new trajectory-based quantum dynamics approaches derived from the (weighted) constraint phase space representation are useful and practical for describing dynamical processes of composite quantum systems in gas phase as well as in condensed phase.

preprint2022arXiv

Unified Formulation of Phase Space Mapping Approaches for Nonadiabatic Quantum Dynamics

Nonadiabatic dynamical processes are one of the most important quantum mechanical phenomena in chemical, materials, biological, and environmental molecular systems, where the coupling between different electronic states is either inherent in the molecular structure or induced by the (intense) external field. The curse of dimensionality indicates the intractable exponential scaling of calculation effort with system size and restricts the implementation of numerically exact approaches for realistic large systems. The phase space formulation of quantum mechanics offers an important theoretical framework for constructing practical approximate trajectory-based methods for quantum dynamics. This Account reviews our recent progress in phase space mapping theory: a unified framework for constructing the mapping Hamiltonian on phase space for coupled F-state systems where the renowned Meyer-Miller Hamiltonian model is a special case, a general phase space formulation of quantum mechanics for nonadiabatic systems where the electronic degrees of freedom are mapped onto constraint space and the nuclear degrees of freedom are mapped onto infinite space, and an isomorphism between the mapping phase space approach for nonadiabatic systems and that for nonequilibrium electron transport processes.

preprint2022arXiv

Virtual Homogeneity Learning: Defending against Data Heterogeneity in Federated Learning

In federated learning (FL), model performance typically suffers from client drift induced by data heterogeneity, and mainstream works focus on correcting client drift. We propose a different approach named virtual homogeneity learning (VHL) to directly "rectify" the data heterogeneity. In particular, VHL conducts FL with a virtual homogeneous dataset crafted to satisfy two conditions: containing no private information and being separable. The virtual dataset can be generated from pure noise shared across clients, aiming to calibrate the features from the heterogeneous clients. Theoretically, we prove that VHL can achieve provable generalization performance on the natural distribution. Empirically, we demonstrate that VHL endows FL with drastically improved convergence speed and generalization performance. VHL is the first attempt towards using a virtual dataset to address data heterogeneity, offering new and effective means to FL.

preprint2021arXiv

Automated Model Design and Benchmarking of 3D Deep Learning Models for COVID-19 Detection with Chest CT Scans

The COVID-19 pandemic has spread globally for several months. Because its transmissibility and high pathogenicity seriously threaten people's lives, it is crucial to accurately and quickly detect COVID-19 infection. Many recent studies have shown that deep learning (DL) based solutions can help detect COVID-19 based on chest CT scans. However, most existing work focuses on 2D datasets, which may result in low quality models as the real CT scans are 3D images. Besides, the reported results span a broad spectrum on different datasets with a relatively unfair comparison. In this paper, we first use three state-of-the-art 3D models (ResNet3D101, DenseNet3D121, and MC3\_18) to establish the baseline performance on the three publicly available chest CT scan datasets. Then we propose a differentiable neural architecture search (DNAS) framework to automatically search for the 3D DL models for 3D chest CT scans classification with the Gumbel Softmax technique to improve the searching efficiency. We further exploit the Class Activation Mapping (CAM) technique on our models to provide the interpretability of the results. The experimental results show that our automatically searched models (CovidNet3D) outperform the baseline human-designed models on the three datasets with tens of times smaller model size and higher accuracy. Furthermore, the results also verify that CAM can be well applied in CovidNet3D for COVID-19 datasets to provide interpretability for medical diagnosis.

preprint2021arXiv

Efficient kernel-based variable selection with sparsistency

Variable selection is central to high-dimensional data analysis, and various algorithms have been developed. Ideally, a variable selection algorithm shall be flexible, scalable, and with theoretical guarantee, yet most existing algorithms cannot attain these properties at the same time. In this article, a three-step variable selection algorithm is developed, involving kernel-based estimation of the regression function and its gradient functions as well as a hard thresholding. Its key advantage is that it assumes no explicit model assumption, admits general predictor effects, allows for scalable computation, and attains desirable asymptotic sparsistency. The proposed algorithm can be adapted to any reproducing kernel Hilbert space (RKHS) with different kernel functions, and can be extended to interaction selection with slight modification. Its computational cost is only linear in the data dimension, and can be further improved through parallel computing. The sparsistency of the proposed algorithm is established for general RKHS under mild conditions, including linear and Gaussian kernels as special cases. Its effectiveness is also supported by a variety of simulated and real examples.

preprint2021arXiv

The complete forcing numbers of hexagonal systems

Let G be a graph with a perfect matching. A complete forcing set of G is a subset of edges of G to which the restriction of every perfect matching is a forcing set of it. The complete forcing number of G is the minimum cardinality of complete forcing sets of G. Xu et al. gave a characterization for a complete forcing set and derived some explicit formulas for the complete forcing numbers of cata-condensed hexagonal systems. In this paper, we consider general hexagonal systems. We present an upper bound on the complete forcing numbers of hexagonal systems in terms of elementary edge-cut cover and two lower bounds by the number of hexagons and matching number respectively. As applications, we obtain some explicit formulas for the complete forcing numbers of some types of hexagonal systems including parallelogram, regular hexagon- and rectangle-shaped hexagonal systems.

preprint2020arXiv

Computational Study of Defect variant Perovskites A2BX6 for Photovoltaic Applications

A comprehensive study of the structural, electronic, and optical properties of lead-free perovskites has been carried out by means of first principles method based on DFT. The calculations are performed for the compound of the type A2BX6 with A=Rb, and Cs; B=Sn, Pd, and Pt; and X=Cl, Br, and I. The calculated structural parameters (lattice constants and bond lengths) agree well with the experiments. The computed band gap reveals a semiconducting profile for all these compounds showing a decreasing trend of the band gap energy by changing the halide ions consecutively from Cl to Br and Br to I. However, for variation in the B-site cation, the band gap increases by changing the cation from Pd to Pt via Sn. The most likely compounds, Rb2PdBr6 and Cs2PtI6, exhibit a band gap within the optimal range of 0.9-1.6 eV for single-junction photovoltaic applications. The optical properties in terms of the optimal value of the dielectric constant, optical conductivity, and absorption coefficient are also investigated upto the photon energy of 10 eV. Our results indicate that upon changing the halogen ions (Cl by Br and Br by I) the optical properties altered significantly. Maximum dielectric constants and high optical absorption are found for Rb2PdI6 and Cs2PtI6. The unique optoelectronic properties such as ideal band gap, high dielectric constants, and optimum absorption of A2BX6 perovskites could be efficiently utilized in designing high performance single and multi-junction perovskite solar cells.

preprint2020arXiv

Convergence Rates of Inertial Primal-Dual Dynamical Methods for Separable Convex Optimization Problems

In this paper, we propose a second-order continuous primal-dual dynamical system with time-dependent positive damping terms for a separable convex optimization problem with linear equality constraints. By the Lyapunov function approach, we investigate asymptotic properties of the proposed dynamical system as the time $t\to+\infty$. The convergence rates are derived for different choices of the damping coefficients. We also show that the obtained results are robust under external perturbations.

preprint2020arXiv

Scene Text Detection and Recognition: The Deep Learning Era

With the rise and development of deep learning, computer vision has been tremendously transformed and reshaped. As an important research area in computer vision, scene text detection and recognition has been inescapably influenced by this wave of revolution, consequentially entering the era of deep learning. In recent years, the community has witnessed substantial advancements in mindset, approach and performance. This survey is aimed at summarizing and analyzing the major changes and significant progresses of scene text detection and recognition in the deep learning era. Through this article, we devote to: (1) introduce new insights and ideas; (2) highlight recent techniques and benchmarks; (3) look ahead into future trends. Specifically, we will emphasize the dramatic differences brought by deep learning and the grand challenges still remained. We expect that this review paper would serve as a reference book for researchers in this field. Related resources are also collected and compiled in our Github repository: https://github.com/Jyouhou/SceneTextPapers.

preprint2020arXiv

TextSnake: A Flexible Representation for Detecting Text of Arbitrary Shapes

Driven by deep neural networks and large scale datasets, scene text detection methods have progressed substantially over the past years, continuously refreshing the performance records on various standard benchmarks. However, limited by the representations (axis-aligned rectangles, rotated rectangles or quadrangles) adopted to describe text, existing methods may fall short when dealing with much more free-form text instances, such as curved text, which are actually very common in real-world scenarios. To tackle this problem, we propose a more flexible representation for scene text, termed as TextSnake, which is able to effectively represent text instances in horizontal, oriented and curved forms. In TextSnake, a text instance is described as a sequence of ordered, overlapping disks centered at symmetric axes, each of which is associated with potentially variable radius and orientation. Such geometry attributes are estimated via a Fully Convolutional Network (FCN) model. In experiments, the text detector based on TextSnake achieves state-of-the-art or comparable performance on Total-Text and SCUT-CTW1500, the two newly published benchmarks with special emphasis on curved text in natural images, as well as the widely-used datasets ICDAR 2015 and MSRA-TD500. Specifically, TextSnake outperforms the baseline on Total-Text by more than 40% in F-measure.

preprint2019arXiv

Actively tunable terahertz electromagnetically induced transparency analogue based on vanadium-oxide-assisted metamaterials

Recently, phase-change materials (PCMs) have drawn more attention due to the dynamically tunable optical properties. Here, we investigate the active control of electromagnetically induced transparency (EIT) analogue based on terahertz (THz) metamaterials integrated with vanadium oxide (VO2). Utilizing the insulator-to-metal transition of VO2, the amplitude of EIT peak can be actively modulated with a significant modulation depth. Meanwhile the group delay within the transparent window can also be dynamically tuned, achieving the active control of slow light effect. Furthermore, we also introduce independently tunable transparent peaks as well as group delay based on a double-peak EIT with good tuning performance. Finally, based on broadband EIT, the active tuning of quality factor of the EIT peak is also realized. This work introduces active EIT control with more degree of freedom by employing VO2, and can find potential applications in future wireless and ultrafast THz communication systems as multi-channel filters, switches, spacers, logic gates and modulators.

preprint2019arXiv

Coherent vortex dynamics in a strongly-interacting superfluid on a silicon chip

Two-dimensional superfluidity and quantum turbulence are directly connected to the microscopic dynamics of quantized vortices. However, surface effects have prevented direct observations of coherent vortex dynamics in strongly-interacting two-dimensional systems. Here, we overcome this challenge by confining a two-dimensional droplet of superfluid helium at microscale on the atomically-smooth surface of a silicon chip. An on-chip optical microcavity allows laser-initiation of vortex clusters and nondestructive observation of their decay in a single shot. Coherent dynamics dominate, with thermal vortex diffusion suppressed by six orders-of-magnitude. This establishes a new on-chip platform to study emergent phenomena in strongly-interacting superfluids, test astrophysical dynamics such as those in the superfluid core of neutron stars in the laboratory, and construct quantum technologies such as precision inertial sensors.

preprint2019arXiv

Hybridization-induced resonances with high quality factor in a plasmonic concentric ring-disk nanocavity

Plasmonic resonators have drawn more attention due to the ability to confine light into subwavelength scale. However, they always suffer from a low quality (Q) factor owing to the intrinsic loss of metal. Here, we numerically propose a plasmonic resonator with ultra-high Q factor based on plasmonic metal-insulator-metal (MIM) waveguide structures. The resonator consists of a disk cavity surrounded by a concentric ring cavity, possessing an ultra-small volume. Arising from the plasmon hybridization between plasmon modes in the disk and ring cavity, the induced bonding hybridized modes have ultra-narrow full wave at half maximum (FWHM) as well as ultra-high Q factors. The FWHM can be nearly 1 nm and Q factor can be more than 400. Furthermore, such device can act as a refractive index sensor with ultra-high figure of merit (FOM). This work provides a novel approach to design plasmonic high-Q-factor resonators, and has potential on-chip applications such as filters, sensors and nanolasers.

preprint2019arXiv

Strong optical coupling through superfluid Brillouin lasing

Brillouin scattering has applications ranging from signal processing, sensing and microscopy, to quantum information and fundamental science. Most of these applications rely on the electrostrictive interaction between light and phonons. Here we show that in liquids optically-induced surface deformations can provide an alternative and far stronger interaction. This allows the demonstration of ultralow threshold Brillouin lasing and strong phonon-mediated optical coupling for the first time. This form of strong coupling is a key capability for Brillouin-reconfigurable optical switches and circuits, for photonic quantum interfaces, and to generate synthetic electromagnetic fields. While applicable to liquids quite generally, our demonstration uses superfluid helium. Configured as a Brillouin gyroscope this provides the prospect of measuring superfluid circulation with unprecedented precision, and to explore the rich physics of quantum fluid dynamics, from quantized vorticity to quantum turbulence.