			Home About us Contact

Gradient Estimation (gradient + estimation)

Distribution by Scientific Domains

Engineering	50%

Selected Abstracts

Gradient Estimation in Volume Data using 4D Linear Regression

COMPUTER GRAPHICS FORUM, Issue 3 2000
László Neumann
In this paper a new gradient estimation method is presented which is based on linear regression. Previous contextual shading techniques try to fit an approximate function to a set of surface points in the neighborhood of a given voxel. Therefore a system of linear equations has to be solved using the computationally expensive Gaussian elimination. In contrast, our method approximates the density function itself in a local neighborhood with a 3D regression hyperplane. This approach also leads to a system of linear equations but we will show that it can be solved with an efficient convolution. Our method provides at each voxel location the normal vector and the translation of the regression hyperplane which are considered as a gradient and a filtered density value respectively. Therefore this technique can be used for surface smoothing and gradient estimation at the same time. [source]

Resource allocation in satellite networks: certainty equivalent approaches versus sensitivity estimation algorithms

INTERNATIONAL JOURNAL OF COMMUNICATION SYSTEMS, Issue 1 2005
Franco Davoli
Abstract In this paper, we consider a resource allocation problem for a satellite network, where variations of fading conditions are added to those of traffic load. Since the capacity of the system is finite and divided in finite discrete portions, the resource allocation problem reveals to be a discrete stochastic programming one, which is typically NP-hard. We propose a new approach based on the minimization over a discrete constraint set using an estimation of the gradient, obtained through a ,relaxed continuous extension' of the performance measure. The computation of the gradient estimation is based on the infinitesimal perturbation analysis technique, applied on a stochastic fluid model of the network. No closed-forms of the performance measure, nor additional feedback concerning the state of the system, and very mild assumptions on the probabilistic properties about the statistical processes involved in the problem are requested. Such optimization approach is compared with a dynamic programming algorithm that maintains a perfect knowledge about the state of the satellite network (traffic load statistics and fading levels). The comparison shows that the sensitivity estimation capability of the proposed algorithm allows to maintain the optimal resource allocation in dynamic conditions and it is able to provide even better performance than the one reached by employing the dynamic programming approach. Copyright © 2004 John Wiley & Sons, Ltd. [source]

Real-time optimization of dynamic systems using multiple units

INTERNATIONAL JOURNAL OF ROBUST AND NONLINEAR CONTROL, Issue 13 2007
B. Srinivasan
Abstract Model-free, unconstrained, real-time optimization of the operating point of a dynamic system involves forcing the gradient of the cost function to zero. In these methods, gradient estimation is a key issue, for which methods that perturb the input over time are used. The main limitation of these methods is that they require the dynamics of the adaptation to be two orders of magnitude slower than the system dynamics. To circumvent this limitation, a novel, simple, yet effective way of estimating the gradient is presented in this paper. Multiple identical units with non-identical inputs are used and the gradient is computed via finite difference. Thus, the perturbation is along the ,unit dimension', thereby allowing a faster adaptation. The convergence of the scheme is rigorously established via Lyapunov analysis. An illustrative example is provided where the proposed scheme resulted in an 100-fold improvement in the time needed for convergence. Copyright © 2007 John Wiley & Sons, Ltd. [source]