arXiv subjects

Bin Xie

Publications and source records attributed to Bin Xie.

At least 55 records · Page 3Linked to original sources

arXiv2023

A Survey of Robustness and Safety of 2D and 3D Deep Learning Models Against Adversarial Attacks

Benefiting from the rapid development of deep learning, 2D and 3D computer vision applications are deployed in many safe-critical systems, such as autopilot and identity authentication. However, deep learning models are not trustworthy enough because of their limited robustness against adversarial attacks. The physically realizable adversarial attacks further pose fatal threats to the application and human safety. Lots of papers have emerged to investigate the robustness and safety of deep learning models against adversarial attacks. To lead to trustworthy AI, we first construct a general threat model from different perspectives and then comprehensively review the latest progress of both 2D and 3D adversarial attacks. We extend the concept of adversarial examples beyond imperceptive perturbations and collate over 170 papers to give an overview of deep learning model robustness against various adversarial attacks. To the best of our knowledge, we are the first to systematically investigate adversarial attacks for 3D models, a flourishing field applied to many real-world applications. In addition, we examine physical adversarial attacks that lead to safety violations. Last but not least, we summarize present popular topics, give insights on challenges, and shed light on future research on trustworthy AI.

cs.LG↗

arXiv2023

MVMR-FS : Non-parametric feature selection algorithm based on Maximum inter-class Variation and Minimum Redundancy

How to accurately measure the relevance and redundancy of features is an age-old challenge in the field of feature selection. However, existing filter-based feature selection methods cannot directly measure redundancy for continuous data. In addition, most methods rely on manually specifying the number of features, which may introduce errors in the absence of expert knowledge. In this paper, we propose a non-parametric feature selection algorithm based on maximum inter-class variation and minimum redundancy, abbreviated as MVMR-FS. We first introduce supervised and unsupervised kernel density estimation on the features to capture their similarities and differences in inter-class and overall distributions. Subsequently, we present the criteria for maximum inter-class variation and minimum redundancy (MVMR), wherein the inter-class probability distributions are employed to reflect feature relevance and the distances between overall probability distributions are used to quantify redundancy. Finally, we employ an AGA to search for the feature subset that minimizes the MVMR. Compared with ten state-of-the-art methods, MVMR-FS achieves the highest average accuracy and improves the accuracy by 5% to 11%.

cs.LG↗

arXiv2023

Post-training Quantization on Diffusion Models

Denoising diffusion (score-based) generative models have recently achieved significant accomplishments in generating realistic and diverse data. These approaches define a forward diffusion process for transforming data into noise and a backward denoising process for sampling data from noise. Unfortunately, the generation process of current denoising diffusion models is notoriously slow due to the lengthy iterative noise estimations, which rely on cumbersome neural networks. It prevents the diffusion models from being widely deployed, especially on edge devices. Previous works accelerate the generation process of diffusion model (DM) via finding shorter yet effective sampling trajectories. However, they overlook the cost of noise estimation with a heavy network in every iteration. In this work, we accelerate generation from the perspective of compressing the noise estimation network. Due to the difficulty of retraining DMs, we exclude mainstream training-aware compression paradigms and introduce post-training quantization (PTQ) into DM acceleration. However, the output distributions of noise estimation networks change with time-step, making previous PTQ methods fail in DMs since they are designed for single-time step scenarios. To devise a DM-specific PTQ method, we explore PTQ on DM in three aspects: quantized operations, calibration dataset, and calibration metric. We summarize and use several observations derived from all-inclusive investigations to formulate our method, which especially targets the unique multi-time-step structure of DMs. Experimentally, our method can directly quantize full-precision DMs into 8-bit models while maintaining or even improving their performance in a training-free manner. Importantly, our method can serve as a plug-and-play module on other fast-sampling methods, e.g., DDIM. The code is available at https://github.com/42Shawn/PTQ4DM .

cs.CV↗

arXiv2022

MLP-GAN for Brain Vessel Image Segmentation

Brain vessel image segmentation can be used as a promising biomarker for better prevention and treatment of different diseases. One successful approach is to consider the segmentation as an image-to-image translation task and perform a conditional Generative Adversarial Network (cGAN) to learn a transformation between two distributions. In this paper, we present a novel multi-view approach, MLP-GAN, which splits a 3D volumetric brain vessel image into three different dimensional 2D images (i.e., sagittal, coronal, axial) and then feed them into three different 2D cGANs. The proposed MLP-GAN not only alleviates the memory issue which exists in the original 3D neural networks but also retains 3D spatial information. Specifically, we utilize U-Net as the backbone for our generator and redesign the pattern of skip connection integrated with the MLP-Mixer which has attracted lots of attention recently. Our model obtains the ability to capture cross-patch information to learn global information with the MLP-Mixer. Extensive experiments are performed on the public brain vessel dataset that show our MLP-GAN outperforms other state-of-the-art methods. We release our code at https://github.com/bxie9/MLP-GAN

eess.IV↗

arXiv2021

Deep Filtering with DNN, CNN and RNN

This paper is about a deep learning approach for linear and nonlinear filtering. The idea is to train a neural network with Monte Carlo samples generated from a nominal dynamic model. Then the network weights are applied to Monte Carlo samples from an actual dynamic model. A main focus of this paper is on the deep filters with three major neural network architectures (DNN, CNN, RNN). Our deep filter compares favorably to the traditional Kalman filter in linear cases and outperform the extended Kalman filter in nonlinear cases. Then a switching model with jumps is studied to show the adaptiveness and power of our deep filtering. Among the three major NNs, the CNN outperform the others on average. while the RNN does not seem to be suitable for the filtering problem. One advantage of the deep filter is its robustness when the nominal model and actual model differ. The other advantage of deep filtering is real data can be used directly to train the deep neutral network. Therefore, model calibration can be by-passed all together.

eess.SP↗

arXiv2021

The Parameter Sensitivities of a Jump-diffusion Process in Basic Credit Risk Analysis

We detect the parameter sensitivities of bond pricing which is driven by a Brownian motion and a compound Poisson process as the discontinuous case in credit risk research. The strict mathematical deductions are given theoretically due to the explicit call price formula. Furthermore, we illustrate Matlab simulation to verify these conclusions.

q-fin.MF↗

arXiv2021

Pricing S&P 500 Index Options with Lévy Jumps

We analyze various jumps for Heston model, non-IID model and three Lévy jump models for S&P 500 index options. The Lévy jump for the S&P 500 index options is inevitable from empirical studies. We estimate parameters from in-sample pricing through SSE for the BS, SV, SVJ, non-IID and Lévy (GH, NIG, CGMY) models by the method of Bakshi et al. (1997), and utilize them for out-of-sample pricing and compare these models. The sensitivities of the call option pricing for the Lévy models with respect to parameters are presented. Empirically, we show that the NIG model, SV and SVJ models with estimated volatilities outperform other models for both in-sample and out-of-sample periods. Using the in-sample optimized parameters, we find that the NIG model has the least SSE and outperforms the rest models on one-day prediction.

q-fin.MF↗

arXiv2021

Global solvability and convergence to stationary solutions in singular quasilinear stochastic PDEs

We consider singular quasilinear stochastic partial differential equations (SPDEs) studied in \cite{FHSX}, which are defined in paracontrolled sense. The main aim of the present article is to establish the global-in-time solvability for a particular class of SPDEs with origin in particle systems and, under a certain additional condition on the noise, prove the convergence of the solutions to stationary solutions as $t\to\infty$. We apply the method of energy inequality and Poincaré inequality. It is essential that the Poincaré constant can be taken uniformly in an approximating sequence of the noise. We also use the continuity of the solutions in the enhanced noise, initial values and coefficients of the equation, which we prove in this article for general SPDEs discussed in \cite{FHSX} except that in the enhanced noise. Moreover, we apply the initial layer property of improving regularity of the solutions in a short time.

math.PR↗

arXiv2021

THINC scaling method that bridges VOF and level set schemes

We present a novel interface-capturing scheme, THINC-scaling, to unify the VOF (volume of fluid) and the level set methods, which have been developed as two different approaches widely used in various applications. The key to success is to maintain a high-quality THINC reconstruction function using the level set field to accurately retrieve geometrical information and the VOF field to fulfill numerical conservativeness. The interface is well defined as a surface in form of a high-order polynomial, so-called the polynomial surface of interface (PSI). The THINC reconstruction function is then used to update the VOF field via a finite volume method, and the level set field via a semi-Lagrangian method. Seeing the VOF field and the level set field as two different aspects of the THINC reconstruction function, the THINC-scaling scheme preserves at the same time the advantages of both VOF and level set methods, i.e. the mass/volume conservation of the VOF method and the geometrical faithfulness of the level set method, through a straightforward solution procedure. The THINC-scaling scheme allows to represent an interface with high-order polynomials and has algorithmic simplicity which largely eases its implementation in unstructured grids. Two and three dimensional algorithms in both structured and unstructured grids have been developed and verified. The numerical results reveal that the THINC-scaling scheme, as an interface capturing method, is able to provide high-fidelity solution comparable to other most advanced methods, and more profoundly it can resolve sub-grid filament structures if the interface is represented by a polynomial higher than second order.

math.NA↗

arXiv2021

Ergodicity of stochastic Cahn-Hilliard equations with logarithmic potentials driven by degenerate or nondegenerate noises

We study the asymptotic properties of the stochastic Cahn-Hilliard equation with the logarithmic free energy by establishing different dimension-free Harnack inequalities according to various kinds of noises. The main characteristics of this equation are the singularities of the logarithmic free energy at 1 and --1 and the conservation of the mass of the solution in its spatial variable. Both the space-time colored noise and the space-time white noise are considered. For the highly degenerate space-time colored noise, the asymptotic log-Harnack inequality is established under the so-called essentially elliptic conditions. And the Harnack inequality with power is established for non-degenerate space-time white noise.

math.AP↗

arXiv2020

Log-Harnack inequality for reflected SPDEs driven by multiplicative noises and its applications

We mainly investigate the log-Harnack inequality for the reflected stochastic partial differential equation driven by multiplicative noises based on the gradient estimate of the associated Markov semigroup. To do it, the penalization method and the comparison principle for stochastic partial differential equations are adopted. As its applications, we also establish the entropy-cost inequality and study some important estimates of the transition density relative to its invariant measure.

math.PR↗

arXiv2020

Asymptotics of PDE in random environment by paracontrolled calculus

We apply the paracontrolled calculus to study the asymptotic behavior of a certain quasilinear PDE with smeared mild noise, which originally appears as the space-time scaling limit of a particle system in random environment on one dimensional discrete lattice. We establish the convergence result and show a local in time well-posedness of the limit stochastic PDE with spatial white noise. It turns out that our limit stochastic PDE does not require any renormalization. We also show a comparison theorem for the limit equation.

math.PR↗

arXiv2020

Discontinuity-resolving shock-capturing schemes on unstructured grids

Solving compressible flows containing discontinuities remains a major challenge for numerical methods especially on unstructured grids. Thus in this work, we make contributions to shock capturing schemes on unstructured grids with aim of resolving discontinuities with low numerical dissipation. Different from conventional shock capturing schemes which only use polynomials as interpolation functions on unstructured grids, the proposed scheme employs the linear polynomial as well as non-polynomial as reconstruction candidates. For linear polynomial, the second order MUSCL scheme with the MLP (Multi-dimensional Limiting Process) slope limiter is adopted. The multi-dimensional THINC (Tangent of Hyperbola for INterface Capturing) function with quadratic surface representation and Gaussian quadrature, so-called THINC/QQ, is used as the non-polynomial reconstruction candidate. With these reconstruction candidates, a multi-stage boundary variation diminishing (BVD) algorithm which aims to minimize numerical dissipation is designed on unstructured grids to select the final reconstruction function. The resulted shock capturing scheme is named as MUSCL-THINC/QQ-BVD. The performance of the proposed scheme is demonstrated through solving compressible single-phase and multi-phase problems where the discontinuity is the typical flow structure. The numerical results show that the proposed scheme is capable of capturing sharp discontinuous profiles without numerical oscillations as well as resolving vortices associated with Kelvin-Helmholtz instabilities along shear layers and material interfaces. In comparison with schemes only replying on high order polynomials, the proposed scheme shows significant improvement of resolution across discontinuities. Thus, this work provides an accurate and robust shock-capturing scheme to resolve discontinuities in compressible flows.

physics.comp-ph↗

arXiv2019

THINC-scaling scheme that unifies VOF and level set methods

We present a novel interface-capturing scheme, THINC-scaling, to unify the VOF (volume of fluid) and the level set methods, which have been developed as two completely different approaches widely used in various applications. The THINC-scaling scheme preserves at the samectime the advantages of both VOF and level set methods, i.e. the mass/volume conservation of the VOF method and the geometrical faithfulness of the level set method. THINC-scaling scheme allows to represent interface with high-order polynomials, and has algorithmic simplicity which eases its implementation in unstructured grids.

math.NA↗

arXiv2018

High-fidelity simulation of regular waves based on multi-moment finite volume formulation and THINC method

The performance of interFoam (a widely used solver within OpenFOAM package) in simulating the propagation of water waves has been reported to be sensitive to the temporal and spatial resolution. To facilitate more accurate simulations, a numerical wave tank is built based on a high-order accurate Navier-Stokes model, which employs the VPM (volume-average/point-value multi-moment) scheme as the fluid solver and the THINC/QQ method (THINC method with quadratic surface representation and Gaussian quadrature) for the free-surface capturing. Simulations of regular waves in an intermediate water depth are conducted and the results are assessed via comparing with the analytical solutions. The performance of the present model and interFoam solver in simulating the wave propagation is systematically compared in this work. The results clearly demonstrate that compared with interFoam solver, the present model significantly improves the dissipation properties of the propagating wave, where the waveforms as well as the velocity distribution can be substantially maintained while the waves propagating over long distances even with large time steps and coarse grids. It is also shown that the present model requires much less computation time to reach a given error level in comparison with interFoam solver.

physics.flu-dyn↗

arXiv2017

A new shock-capturing scheme for stiff detonation waves problems

A new approach to prevent spurious behavior caused by conventional shock-capturing schemes when solving stiff detonation waves problems is introduced in the present work. Due to smearing of discontinuous solution by the excessive numerical dissipation, conventional shock-capturing schemes have difficulties to obtain the correct location of detonation front without enough grids resolution. To overcome the excessive numerical errors around discontinuities by traditional discretized schemes used in non-reacting high speed compressible flow, we introduce a new shock-capturing scheme in which besides linear function constructed by MUSCL (Monotone Upstream-centered Schemes for Conservation Law) scheme, a step like THINC (Tangent of Hyperbola for INterface Capturing) function is also employed as another candidate in the reconstruction process. The final reconstruction function is determined by boundary variation diminishing (BVD) algorithm by which numerical dissipation around discontinuities can be reduced significantly. The new resulted shock-capturing scheme is named MUSCL-THINC-BVD. One- and two-dimensional comparative numerical tests about stiff detonation waves problems are conducted with the 5th order WENO (Weighted Essentially Non-Oscillatory) and MUSCL-THINC-BVD scheme respectively, which show MUSCL-THINC-BVD scheme can capture the correct position of detonation waves with improved resolution while WENO scheme, in spite of higher order, produces spurious waves. Compared with other existing methods which involves extra treatments by accepting the smeared out discontinuities profiles, the current method obtain the correct but also sharp detonation front by fundamentally reducing numerical dissipation errors from shock-capturing schemes. Thus the proposed approach is an effective but simple method to solve stiff detonation problems.

physics.comp-ph↗

arXiv2017

Some practical versions of boundary variation diminishing (BVD) algorithm

This short note presents some variant schemes of boundary variation diminishing (BVD) algorithm in one dimension with the results of numerical tests for linear advection equation to facilitate practical use. In spite of being presented in 1D fashion, all the schemes are simple and easy to implement in multi-dimensions on structured and unstructured grids for nonlinear and system equations.

math.NA↗

arXiv2017

Implementation of BVD (boundary variation diminishing) algorithm in simulations of compressible multiphase flows

We present in this work a new reconstruction scheme, so-called MUSCL-THINC-BVD scheme, to solve the five-equation model for interfacial two phase flows. This scheme employs the traditional shock capturing MUSCL (Monotone Upstream-centered Schemes for Conservation Law) scheme as well as the interface sharpening THINC (Tangent of Hyperbola for INterface Capturing) scheme as two building-blocks of spatial reconstruction using the BVD (boundary variation diminishing) principle that minimizes the variations (jumps) of the reconstructed variables at cell boundaries, and thus effectively reduces the numerical dissipations in numerical solutions. The MUSCL-THINC-BVD scheme is implemented to all state variables and volume fraction, which realizes the consistency among volume fraction and other physical variables. Benchmark tests are carried out to verify the capability of the present method in capturing the material interface as a well-defined sharp jump in volume fraction, as well as significant improvement in solution quality. The proposed scheme is a simple and effective method of practical significance for simulating compressible interfacial multiphase flows.

physics.comp-ph↗