Catalogo Articoli (Spogli Riviste)
OPAC HELP
Titolo: Linear concepts and hidden variables
Autore: Grove, AJ; Roth, D;
 Indirizzi:
 NECI, Princeton, NJ USA NECI Princeton NJ USANECI, Princeton, NJ USA Univ Illinois, Dept Comp Sci, Urbana, IL 61801 USA Univ Illinois Urbana IL USA 61801 is, Dept Comp Sci, Urbana, IL 61801 USA
 Titolo Testata:
 MACHINE LEARNING
fascicolo: 12,
volume: 42,
anno: 2001,
pagine: 123  141
 SICI:
 08856125(200101)42:12<123:LCAHV>2.0.ZU;22
 Fonte:
 ISI
 Lingua:
 ENG
 Keywords:
 linear functions; Winnow; expectationmaximization; Naire Bayes;
 Tipo documento:
 Article
 Natura:
 Periodico
 Settore Disciplinare:
 Engineering, Computing & Technology
 Citazioni:
 15
 Recensione:
 Indirizzi per estratti:
 Indirizzo: Grove, AJ 44 Murray Pl, Princeton, NJ 08540 USA 44 Murray Pl Princeton NJUSA 08540 Pl, Princeton, NJ 08540 USA



 Citazione:
 A.J. Grove e D. Roth, "Linear concepts and hidden variables", MACH LEARN, 42(12), 2001, pp. 123141
Abstract
We study a learning problem which allows for a "fair" comparison between unsupervised learning methodsprobabilistic model construction, and more traditional algorithms that directly learn a classification. The merits of each approach are intuitively clear: inducing a model is more expensive computationally, but may support a wider range of predictions. Its performance, however, will depend on how well the postulated probabilistic model fits that data. To compare the paradigms we consider a model which postulates a single binaryvalued hidden variable on which all other attributes depend. In this model, finding the most likely value of any one variable (given known values for the others) reduces to testing a linear function of the observedvalues. We learn the model with two techniques: the standard EM algorithm,and a new algorithm we develop based on covariances. We compare these, in a controlled fashion, against an algorithm (a version of Winnow) that attempts to find a good linear classifier directly. Our conclusions help delimitthe fragility of using a model that is even "slightly" simpler than the distribution actually generating the data, vs. the relative robustness of directly searching for a good predictor.
ASDD Area Sistemi Dipartimentali e Documentali, Università di Bologna, Catalogo delle riviste ed altri periodici
Documento generato il 05/04/20 alle ore 12:14:14