With the rapid developments of artificial intelligence(AI)and deep learning(DL)techniques,it is critical to ensure the security and robustness of the deployed algorithms.Recently,the security vulnerability of DL algor...With the rapid developments of artificial intelligence(AI)and deep learning(DL)techniques,it is critical to ensure the security and robustness of the deployed algorithms.Recently,the security vulnerability of DL algorithms to adversarial samples has been widely recognized.The fabricated samples can lead to various misbehaviors of the DL models while being perceived as benign by humans.Successful implementations of adversarial attacks in real physical-world scenarios further demonstrate their practicality.Hence,adversarial attack and defense techniques have attracted increasing attention from both machine learning and security communities and have become a hot research topic in recent years.In this paper,we first introduce the theoretical foundations,algorithms,and applications of adversarial attack techniques.We then describe a few research efforts on the defense techniques,which cover the broad frontier in the field.Several open problems and challenges are subsequently discussed,which we hope will provoke further research efforts in this critical area.展开更多
This paper develops and analyzes a stochastic derivative-free optimization strategy.A key feature is the state-dependent adaptive variance.We prove global convergence in probability with algebraic rate and give the qu...This paper develops and analyzes a stochastic derivative-free optimization strategy.A key feature is the state-dependent adaptive variance.We prove global convergence in probability with algebraic rate and give the quantitative results in numerical examples.A striking fact is that convergence is achieved without explicit information of the gradient and even without comparing different objective function values as in established methods such as the simplex method and simulated annealing.It can otherwise be compared to annealing with state-dependent temperature.展开更多
City outskirts serve as the concentration centers for the pollutants discharged from various sources such as industry,agriculture and transportation.With the acceleration of industrialization and urbanization,the eco-...City outskirts serve as the concentration centers for the pollutants discharged from various sources such as industry,agriculture and transportation.With the acceleration of industrialization and urbanization,the eco-environment of the city outskirts has become a hot spot of public concern.An analysis was conducted in this paper on the soil in the outskirts of Yixing City using frequency distribution functions.The heavy metal sources in this region had been divided into two components,i.e.natural background component and strong human disturbance component.The corresponding interpretation of the distribution pattern and features of heavy metal elements was presented by spatial analysis.The results showed that the strong human disturbance components of Hg,Pb,and Cu accounted for 36.9%, 26.7%,and 23.3%in their contents respectively,which indicated directly the serious effect of human activities on heavy metal contents.Hg and Pb,because of the human disturbance,showed the greatest spatial variability,and human activities intensified the heterogeneity of the spatial distribution.The anisotropic analysis showed the higher organisation of Hg and Zn in the direction of urban-rural transition,which indicated their spatial characteristics with urban-rural transition.Pb displayed distinctive structure in the vertical direction of urban-rural transition,which was largely controlled by highway distribution.Cu content in paddy fields was significantly higher than those of other land-use-types,and the agricultural non-point source pollution played an important role in the distribution pattern of Cu.展开更多
The objective of this paper is to review recent developments in numerical reconstruction methods for inverse transport problems in imaging applications,mainly optical tomography,fluorescence tomography and bioluminesc...The objective of this paper is to review recent developments in numerical reconstruction methods for inverse transport problems in imaging applications,mainly optical tomography,fluorescence tomography and bioluminescence tomography.In those inverse problems,one aims at reconstructing physical parameters,such as the absorption coefficient,the scattering coefficient and the fluorescence light source,inside heterogeneous media,from partial knowledge of transport solutions on the boundaries of the media.The physical parameters recovered can be used for diagnostic purpose.Numerical reconstruction techniques for those inverse transport problems can be roughly classified into two categories:linear reconstruction methods and nonlinear reconstruction methods.In the first type of methods,the inverse problems are linearized around some known background to obtain linear inverse problems.Classical regularization techniques are then applied to solve those inverse problems.The second type of methods are either based on regularized nonlinear least-square techniques or based on gradient-driven iterative methods for nonlinear operator equations.In either case,the unknown parameters are iteratively updated until the solutions of the transport equations with the those parameters match the measurements to a certain extent.We review linear and nonlinear reconstruction methods for inverse transport problems in medical imaging with stationary,frequency-domain and time-dependent data.The materials presented include both existing and new results.Meanwhile,we attempt to present similar algorithms for different problems in the same framework to make it more straightforward to generalize those algorithms to other inverse(transport)problems.展开更多
基金Ant Financial,Zhejiang University Financial Technology Research Center.
文摘With the rapid developments of artificial intelligence(AI)and deep learning(DL)techniques,it is critical to ensure the security and robustness of the deployed algorithms.Recently,the security vulnerability of DL algorithms to adversarial samples has been widely recognized.The fabricated samples can lead to various misbehaviors of the DL models while being perceived as benign by humans.Successful implementations of adversarial attacks in real physical-world scenarios further demonstrate their practicality.Hence,adversarial attack and defense techniques have attracted increasing attention from both machine learning and security communities and have become a hot research topic in recent years.In this paper,we first introduce the theoretical foundations,algorithms,and applications of adversarial attack techniques.We then describe a few research efforts on the defense techniques,which cover the broad frontier in the field.Several open problems and challenges are subsequently discussed,which we hope will provoke further research efforts in this critical area.
基金This work was supported in part by the Research Grants Council of Hong Kong （CityU 11276816, CityU 11212717, and CityU C1008-16G）, the Innovation and Technology Commission of Hong Kong （ITS/168/17）, and the National Natural Science Foundation of China （61572412 and 61772236）.
基金This work was partially supported by the National Natural Science Foundation of China （U1636215, 61572492, 61650202, 61772236, and 61372191） and the National Key Research and Development Program （2016YFB0800802, 2016YFB0800803, 2016YFB0800804, 2017YFB0802204, 2016QY03D0601, 2016QY03D0603, and 2016YFB0800303）.
基金partially supported by the National Science Foundation through grants DMS-2208504(BE),DMS-1913309(KR),DMS-1937254(KR),and DMS-1913129(YY)support from Dr.Max Rossler,the Walter Haefner Foundation,and the ETH Zurich Foundation.
文摘This paper develops and analyzes a stochastic derivative-free optimization strategy.A key feature is the state-dependent adaptive variance.We prove global convergence in probability with algebraic rate and give the quantitative results in numerical examples.A striking fact is that convergence is achieved without explicit information of the gradient and even without comparing different objective function values as in established methods such as the simplex method and simulated annealing.It can otherwise be compared to annealing with state-dependent temperature.
基金Supported by the Key Project of the National Natural Science Foundation of China(Grant No.90411015)the National Basic Research Program of China(Grant No.2002CB410810)
文摘City outskirts serve as the concentration centers for the pollutants discharged from various sources such as industry,agriculture and transportation.With the acceleration of industrialization and urbanization,the eco-environment of the city outskirts has become a hot spot of public concern.An analysis was conducted in this paper on the soil in the outskirts of Yixing City using frequency distribution functions.The heavy metal sources in this region had been divided into two components,i.e.natural background component and strong human disturbance component.The corresponding interpretation of the distribution pattern and features of heavy metal elements was presented by spatial analysis.The results showed that the strong human disturbance components of Hg,Pb,and Cu accounted for 36.9%, 26.7%,and 23.3%in their contents respectively,which indicated directly the serious effect of human activities on heavy metal contents.Hg and Pb,because of the human disturbance,showed the greatest spatial variability,and human activities intensified the heterogeneity of the spatial distribution.The anisotropic analysis showed the higher organisation of Hg and Zn in the direction of urban-rural transition,which indicated their spatial characteristics with urban-rural transition.Pb displayed distinctive structure in the vertical direction of urban-rural transition,which was largely controlled by highway distribution.Cu content in paddy fields was significantly higher than those of other land-use-types,and the agricultural non-point source pollution played an important role in the distribution pattern of Cu.
基金Project supported by the National Natural Science Foundation of China(Nos.61472357 and 61571063)the Open Fund of State Key Laboratory of Cryptology+1 种基金the Major Scientific Research Project of Zhejiang Labthe Alibaba-Zhejiang University Joint Institute of Frontier Technologies
基金partially supported by National Science Foundation(NSF)through grant DMS-0914825a faculty development award from the University of Texas at Austin。
文摘The objective of this paper is to review recent developments in numerical reconstruction methods for inverse transport problems in imaging applications,mainly optical tomography,fluorescence tomography and bioluminescence tomography.In those inverse problems,one aims at reconstructing physical parameters,such as the absorption coefficient,the scattering coefficient and the fluorescence light source,inside heterogeneous media,from partial knowledge of transport solutions on the boundaries of the media.The physical parameters recovered can be used for diagnostic purpose.Numerical reconstruction techniques for those inverse transport problems can be roughly classified into two categories:linear reconstruction methods and nonlinear reconstruction methods.In the first type of methods,the inverse problems are linearized around some known background to obtain linear inverse problems.Classical regularization techniques are then applied to solve those inverse problems.The second type of methods are either based on regularized nonlinear least-square techniques or based on gradient-driven iterative methods for nonlinear operator equations.In either case,the unknown parameters are iteratively updated until the solutions of the transport equations with the those parameters match the measurements to a certain extent.We review linear and nonlinear reconstruction methods for inverse transport problems in medical imaging with stationary,frequency-domain and time-dependent data.The materials presented include both existing and new results.Meanwhile,we attempt to present similar algorithms for different problems in the same framework to make it more straightforward to generalize those algorithms to other inverse(transport)problems.
