D. J. Hand (auth.), P. Cheeseman, R. W. Oldford (eds.)'s Selecting Models from Data: Artificial Intelligence and PDF

By D. J. Hand (auth.), P. Cheeseman, R. W. Oldford (eds.)

ISBN-10: 0387942815

ISBN-13: 9780387942810

ISBN-10: 1461226600

ISBN-13: 9781461226604

This quantity is a range of papers awarded on the Fourth foreign Workshop on man made Intelligence and data held in January 1993. those biennial workshops have succeeded in bringing jointly researchers from man made Intelligence and from records to debate difficulties of mutual curiosity. The alternate has broadened study in either fields and has strongly encour­ elderly interdisciplinary paintings. The subject ofthe 1993 AI and facts workshop was once: "Selecting types from Data". The papers during this quantity attest to the range of methods to version choice and to the ubiquity of the matter. either facts and synthetic intelligence have independently built methods to version choice and the corresponding algorithms to enforce them. yet as those papers clarify, there's a excessive measure of overlap among the various ways. specifically, there's contract that the basic challenge is the avoidence of "overfitting"-Le., the place a version matches the given info very heavily, yet is a negative predictor for brand new information; in different phrases, the version has partially outfitted the "noise" within the unique data.

Show description

Read Online or Download Selecting Models from Data: Artificial Intelligence and Statistics IV PDF

Similar statistics books

Methods and Applications of Statistics in Clinical Trials, - download pdf or read online

Tools and functions of information in scientific Trials, quantity 2: making plans, research, and Inferential equipment comprises updates of verified literature from the Wiley Encyclopedia of scientific Trials in addition to unique fabric in keeping with the newest advancements in medical trials. ready by means of a number one professional, the second one quantity comprises a number of contributions from present widespread specialists within the box of clinical examine.

The Elements of Statistical Learning: Data Mining, Inference by Trevor Hastie PDF

In the past decade there was an explosion in computation and knowledge know-how. With it have come huge quantities of knowledge in quite a few fields equivalent to drugs, biology, finance, and advertising. The problem of realizing those info has resulted in the advance of recent instruments within the box of data, and spawned new components equivalent to info mining, laptop studying, and bioinformatics.

Download e-book for iPad: Economics (Barron's Business Review Series) by Walter J. Wessels

Books in Barron's "Business assessment sequence" are meant typically for lecture room use. They make first-class vitamins to major texts whilst incorporated in college-level enterprise classes. In grownup schooling and company brush-up courses they could function major textbooks. All titles during this sequence comprise evaluate questions with solutions.

Download e-book for kindle: Probability, Statistics and Time: A collection of essays by M. S. Bartlett F.R.S. (auth.)

A few years in the past while I. assembled a few basic articles and lectures on likelihood and records, their e-book (Essays in chance and facts, Methuen, London, 1962) bought a a few­ what larger reception than I have been ended in anticipate of the sort of miscellany. i'm for this reason tempted to danger publishing this moment assortment, the identify i've got given it (taken from the 1st lecture) seeming to me to point a coherence in my articles which my publishers may possibly rather be susceptible to question.

Additional resources for Selecting Models from Data: Artificial Intelligence and Statistics IV

Sample text

See Bhansali & Downham (1977), Atkinson (1980, 1981) and Friedman (1991). Critics of such an argument are concerned that such ad hoc choice of A lacks theoretical justification. They argue that any finite Awould lead to an inconsistent criterion. Hence a penalty term that depends on the sample size should be used. Specifically, it has been suggested that A = An should satisfy An --+ 00 and An/n --+ O. See Bozdogan (1987), Nishii (1988) and Zhao et al. (1986a,b). The purpose of this article is to provide a unified approach to the assessment of criterion (1).

Similarly, all distributions q( z) whose marginal is q(y) can be viewed as a manifold. We call it the observed manifold. The E-step of EM algorithm can be viewed as projection onto the observed manifold, and the M-step as projection onto the model manifold. The minimum distance of the two manifolds measured by the K-L divergence is achieved by the estimated empirical distribution q( z) and the estimated true distribution p( z IfJ). 4 Criterion for Model Selection Let us consider criteria to choose good models from a set of candidate models {pi(zI8 i )} on the basis of the observed data q(y).

An information-theoretic approach to model selection, through minimum-bit data representation, is explored. Similarity of the asymptotic form of Rissanen's criterion, obtained from a minimum-bit data representation approach, to criteria derived from a Bayesian approach, is discussed. KEY WORDS: model selection, model evaluation, Akaike's information criterion, Ale, Schwarz's criterion, Kashyap's criterion, Bayesian inference, posterior probabilities. 4,1 Introduction; sets of alternative models Usually a statistical analysis involves consideration of a set of alternative models for the data.

Download PDF sample

Selecting Models from Data: Artificial Intelligence and Statistics IV by D. J. Hand (auth.), P. Cheeseman, R. W. Oldford (eds.)

by Paul

Rated 4.79 of 5 – based on 22 votes