Showing papers on "Interpretability published in 1992"

A common language effect size statistic.

Abstract: Some of the shortcomings in interpretability and generalizability of the effect size statistics currently available to researchers can be overcome by a statistic that expresses how often a score sampled from one distribution will be greater than a score sampled from another distribution. The statistic, the common language effect size indicator, is easily calculated from sample means and variances (or from proportions in the case of nominal-level data). It can be used for expressing the effect observed in both independent and related sample designs and in both 2-group and n-group designs

Measurement of reliability for categorical data in medical research.

Abstract: The problem of measuring reliability of categorical measurements, particularly diagnostic categorizations, is addressed. The approach is based on classical measurement theory and requires interpretability of the reliability coefficients in terms of loss of precision in estimation or power in statistical tests. A general model is proposed, leading to definition of reliability indices. Design and estimation approaches are discussed. Issues and approaches found in the research literature that either lead to confusing or misleading results are presented. The signs and symptoms of unreliable diagnoses are identified, and strategies for improving the reliability of such diagnoses are discussed.

Unifying the landmark developments in optimal bounding ellipsoid identification

Abstract: : A quite general class of Optimal Bounding Ellipsoid (OBE) algorithms including all methods published to date, can be unified into a single framework called the Unified OBE (UOBE) algorithm. UOBE is based on generalized weighted recursive least squares in which very broad classes of 'forgetting factors' and data weights may be employed. Different instances of UOBE are distinguished by their weighting policies and the criteria used to determine their optimal values. A study of existing OBE algorithms, with a particular interest in the tradeoff between algorithm performance interpretability and convergence properties, is presented. Results suggest that an interpretable, converging UOBE algorithm will be found. In this context, a new UOBE technique, the set membership stochastic approximation (SM-SA) algorithm is introduced. SM-SA possesses interpretable optimization measures and known conditions under which its estimator will converge.

Unary interpretability logic

Abstract: Let T be an arithmetical theory. We introduce a unary modal operator T to be interpreted arithmetically as the unary interpretability pred- icate over T. We present complete axiomatizations of the (unary) interpreta- bility principles underlying two important classes of theories. We also prove some basic modal results about these new axiomatizations.

Interpretable Exploratory Projection Pursuit

Abstract: I propose a modification of exploratory projection pursuit which trades accuracy for interpretability in the resulting description. Interpretability, a generalization of parsimony, is based on the ideas of rotation in factor analysis and of entropy. It is defined as the simplicity of the coefficients which specify the description’s projections. A weighted optimization approach similar to roughness penalty curve-fitting is used to search for a more understandable description, with interpretability replacing smoothness. A real data example is presented. The method retains the nonlinear versatility of projection pursuit but has more intuitive appeal.

NES: a neural shell for diagnostic expert systems*

Abstract: We discuss the design choices leading to NES - a totally neural shell for diagnostic expert systems - emphasizing its modularity and the semantic interpretability of most individual neurons. Furthermore, we examine possible limitations of this approach, which suggest the opportunity of integrating NES with connectionist or traditional “symbolic” modules for specific knowledge representation and processing problems. Finally, an application of this system is briefly illustrated.

Probability Plots and Modern Statistical Software

Abstract: Probability plots play a key role in the statistical analysis of one-dimensional samples, emphasizing summarization and visualization. As modern workstations with increased power and agile graphics facilities have become widely available, modern statistical software packages are also increasingly emphasizing visualization and graphical exploratory data analysis. One would expect to see in such packages adequate, if not extensive, implementation of probability plotting methods. Nonetheless, the normal Q-Q plot, being a popular tool for assessing the normality of a sample, is the only type usually found (if at all). In this article we briefly review the basic types of probability plots and we make a suggestion as to what may form the base for the development of a general probability plotting procedure. We propose a variant of the P-P plot, instead of the more popular Q-Q plot, as the former enjoys better interpretability and it is more amenable to generalization. In a workstation based computational environment, the functionality and interpretability of P-P plots are even more amenable to improvement.