摘要 :
Power dissipation has become a significant concern for integrated circuit design in nanometric CMOS technology. To reduce power consumption, approximate implementations of a circuit have been considered as a potential solution for...
展开
Power dissipation has become a significant concern for integrated circuit design in nanometric CMOS technology. To reduce power consumption, approximate implementations of a circuit have been considered as a potential solution for applications in which strict exactness is not required. In approximate computing, power reduction is achieved through the relaxation of the often demanding requirement of accuracy. In this paper, new approximate adders are proposed for low-power imprecise applications by using logic reduction at the gate level as an approach to relaxing numerical accuracy. Transmission gates are utilized in the designs of two approximate full adders with reduced complexity. A further positive feature of the proposed designs is the reduction of the critical path delay. The approximate adders show advantages in terms of power dissipation over accurate and recently proposed approximate adders. An image processing application is presented using the proposed approximate adders to evaluate the efficiency in power and delay at application level.
收起
摘要 :
Power dissipation has become a significant concern for integrated circuit design in nanometric CMOS technology. To reduce power consumption, approximate implementations of a circuit have been considered as a potential solution for...
展开
Power dissipation has become a significant concern for integrated circuit design in nanometric CMOS technology. To reduce power consumption, approximate implementations of a circuit have been considered as a potential solution for applications in which strict exactness is not required. In approximate computing, power reduction is achieved through the relaxation of the often demanding requirement of accuracy. In this paper, new approximate adders are proposed for low-power imprecise applications by using logic reduction at the gate level as an approach to relaxing numerical accuracy. Transmission gates are utilized in the designs of two approximate full adders with reduced complexity. A further positive feature of the proposed designs is the reduction of the critical path delay. The approximate adders show advantages in terms of power dissipation over accurate and recently proposed approximate adders. An image processing application is presented using the proposed approximate adders to evaluate the efficiency in power and delay at application level.
收起
摘要 :
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated...
展开
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated by an FNN, and consequently, the essential approximation ability of an FNN can not been revealed. In this paper, by establishing both upper and lower bound estimations on degree of approximation, the essential approximation ability of a class of FNNs is clarified in terms of the modulus of smoothness of functions to be approximated. The involved FNNs can not only approximate any continuous functions arbitrarily well, but also provide an explicit lower bound on number of hidden units required. By making use of approximation tools, it is shown that when the functions to be approximated are Lipschitzian, the approximation speed of the FNNs is determined by modulus of smoothness of the functions.
收起
摘要 :
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated...
展开
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated by an FNN, and consequently, the essential approximation ability of an FNN can not been revealed. In this paper, by establishing both upper and lower bound estimations on degree of approximation, the essential approximation ability of a class of FNNs is clarified in terms of the modulus of smoothness of functions to be approximated. The involved FNNs can not only approximate any continuous functions arbitrarily well, but also provide an explicit lower bound on number of hidden units required. By making use of approximation tools, it is shown that when the functions to be approximated are Lipschitzian, the approximation speed of the FNNs is determined by modulus of smoothness of the functions.
收起
摘要 :
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated...
展开
There have been various studies on approximation ability of feedforward neural networks (FNNs). Most of the existing studies are, however, only concerned with density or upper bound estimation on how a function can be approximated by an FNN, and consequently, the essential approximation ability of an FNN can not been revealed. In this paper, by establishing both upper and lower bound estimations on degree of approximation, the essential approximation ability of a class of FNNs is clarified in terms of the modulus of smoothness of functions to be approximated. The involved FNNs can not only approximate any continuous functions arbitrarily well, but also provide an explicit lower bound on number of hidden units required. By making use of approximation tools, it is shown that when the functions to be approximated are Lipschitzian, the approximation speed of the FNNs is determined by modulus of smoothness of the functions.
收起
摘要 :
Though quasi-Newton methods have been extensively studied in the literature, they either suffer from local convergence or use a series of line searches for global convergence. In this work, we propose a line search free greedy qua...
展开
Though quasi-Newton methods have been extensively studied in the literature, they either suffer from local convergence or use a series of line searches for global convergence. In this work, we propose a line search free greedy quasi-Newton method with adaptive steps and establish explicit nonasymptotic bounds for both the global convergence rate and local superlinear rate. Our novel idea lies in the design of multiple greedy quasi-Newton updates to control the Hessian approximation error and a simple mechanism to adjust stepsizes to ensure function improvement per iteration. The global superlinear convergence 1 of our method is validated via numerical experiments.
收起
摘要 :
An analysis of online learning for adaptive optimal control through value iteration is presented. Stability of the system operated using any single immature control policy during the learning stage is established. The contribution...
展开
An analysis of online learning for adaptive optimal control through value iteration is presented. Stability of the system operated using any single immature control policy during the learning stage is established. The contribution of this work is incorporating approximation errors present in the function approximation process in developing the results. This is done through establishing sufficient conditions for boundedness of the approximated value function and for stability of the system operated by the control policy during the learning process.
收起
摘要 :
Wideband power spectrum sensing is fundamental for numerous applications. When side information on the potentially active emitters is available, such as carriers and spectral masks, it should be exploited to improve sensing perfor...
展开
Wideband power spectrum sensing is fundamental for numerous applications. When side information on the potentially active emitters is available, such as carriers and spectral masks, it should be exploited to improve sensing performance. Here the power spectrum is modeled as a weighted sum of candidate spectral density primitives. The objective is to estimate the unknown weights from a few randomly filtered broadband power measurement bits, taken using a network of low-end sensors. A linear programming formulation that exploits the sparsity in the unknown weights is proposed. A better approach follows, which exploits the approximately Gaussian distribution of the errors in the power measurements prior to quantization, in a maximum likelihood formulation that includes a sparsity-inducing penalty term. Simulations show that the model weights can be accurately estimated from few bits, even when the errors are significant.
收起
摘要 :
A stable adaptive fuzzy control methods is proposed for multiple-input multiple-output uncertain nonlinear systems. The adaptive law utilizes two type of errors in the adaptive fuzzy systems, the tracking error and approximation e...
展开
A stable adaptive fuzzy control methods is proposed for multiple-input multiple-output uncertain nonlinear systems. The adaptive law utilizes two type of errors in the adaptive fuzzy systems, the tracking error and approximation error. This control law is applied to a two link robot manipulator, and simulation results show its validity.
收起
摘要 :
A rigorous analysis of the approximation error in lightly damped systems is given. Easy-to-check conditions under which neglecting the off-diagonal elements of the normalized damping matrix can result in large approximation errors are presented.
原文获取