Showing papers on "Robustness (computer science) published in 1972"

The 1972 Wald Lecture Robust Statistics: A Review

Abstract: This is a selective review on robust statistics, centering on estimates of location, but extending into other estimation and testing problems. After some historical remarks, several possible concepts of robustness are critically reviewed. Three important classes of estimates are singled out and some basic heuristic tools for assessing properties of robust estimates (or test statistics) are discussed: influence curve, jackknifing. Then we give some asymptotic and finite sample minimax results for estimation and testing. The material is complemented by miscellaneous remarks on: computational aspects; other estimates; scale, regression, time series and other estimation problems; some tentative practical recommendations.

Robust estimation of signal amplitude

Abstract: An iterative estimation procedure involving the use of a light limiter is proposed for obtaining robust estimates of the amplitude of a signal of known form in additive nearly Gaussian noise. In the case of a constant signal the procedure, referred to as the iterative limiter estimator (ILE), is exhibited as a form of stochastic approximation and the procedure is shown to be robust in a well-defined asymptotically min-max sense. The ILE is also shown to be asymptotically robust in terms of efficiency relative to the sample mean. in the case of a time-varying signal the procedure is referred to as an iterative limiter-correlator estimator (ILCE). Simulation results show that both the ILE and the ILCE possess a high degree of robustness for seemingly mild, but potent, deviations from the Gaussian model. This robustness is obtained at the expense of a relatively small loss in efficiency when the noise is actually Gaussian.

Factor Analysis as Numerical Induction: How to Judge a Book by its Cover1

Abstract: The use of some forms of data in factor analysis for purposes of numerical induction has recently been questioned (Sackett 1969). The nature of “qualitative” and “quantitative” data and the assumptions of the factor analysis model are discussed in this paper. Results of a plasmode—a worked example with relatively well understood data—confirm the robustness of factor analysis under varying degrees of violation of the model More and more examples of the usefulness of factor analysis as a general numerical method of induction are accumulating.

Robust filtering for linear systems

Abstract: Recursive robust filtering for a discrete, linear stochastic system with additive white noise disturbances is considered. The initial state and plant disturbances are assumed to be Gaussian and the partial covariance of each measurement over a finite region is assumed bounded from below. A soft limiter and patched-Gaussian density are shown to be the optimal min-max estimator and the least favorable measurement density, respectively. An approximate filter is proposed and an example is given.