Wolpert, D. H. (1995). "The relationship between PAC, the statistical physics framework, the Bayesian framework, and the VC framework." In Wolpert, D. H., editor, The Mathematics of Generalization, Addison-Wesley.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....model given the available data and nature of function. Additional details on ensemble modeling can be found elsewhere (Barai and Reich, 1999) 5 Summary and conclusions In respect to modeling all kinds of problems, no method can be singled out to be superior than their peers (Schaffer, 1994; Wolpert, 1995). In fact, in classification problems, averaged over all modeling problems, all methods are equal. For the modeling of marine propeller behavior, NN are candidates for generating good quality models. Multilayer perceptron neural networks provide means for modeling arbitrary functions. They can ....

Wolpert, D. H. (1995). "The relationship between PAC, the statistical physics framework, the Bayesian framework, and the VC framework." In Wolpert, D. H., editor, The Mathematics of Generalization, Addison-Wesley.

....employ any of the successful itemset generation algorithms [AS94] An itemset with its support provides information on the frequency of a certain pattern and can be thought of as a local descriptor of the data. Since training a classifier can be thought of as finding a condensed model of the data [W94], the question arises whether itemsets can also be used for classification purposes. This paper provides a positive answer by showing: i) how itemsets can be used to construct a partially lazy classifier [A97] and (ii) that the resulting classifier leads to superior accuracy in many cases. In ....

....system [K 94] No discretization was applied for C4.5. Table 1 provides detailed information on the data sets used and summarizes the accuracies achieved by the four algorithms on the 21 data sets. Keeping in mind that no classification method can outperform all others in all possible domains [W94], we can draw several conclusions from Table 1. Firstly, it is clear that TAN and LB are in general ahead of the others in terms of prediction accuracy. This is also supported by the fact that they win in 6 (TAN) and 8 (LB) cases respectively. On the other hand, if the domain concept can naturally ....

D.H.Wolpert, " The relationship between PAC, the statistical physics framework, the Bayesian framework, and the VC framework.", In D.H.Wolpert (ed.), The Mathematics of Generalization, 1994.

....q, m) dh E(C f, h, q) S d P(h d) P(d Y f, d X ) P(d X f, m, q) I do not equate P(d X f, q, m) with P(d X m) as is conventionally (though implicitly) done in most theoretical supervised learning because in general the test set point q may be coupled to d X and even f. See [Wolpert 1995a] 8 III BIAS PLUS VARIANCE FOR QUADRATIC LOSS i) The bias plus variance formula This section reviews the conventional bias plus variance formula for quadratic los, with a fixed targets and averages over training sets. Write E(Y F f, q) S y y f(q, y) E(Y F 2 f, q) S y y 2 ....

....E(C d) We know d, and therefore can fix its value in the conditioning event. We do not know f. Analyzing E(C d) is the purview of Bayesian analysis [Buntine and Weigend 1991, Bernardo and Smith 1994] Generically, it says that for quadratic loss, one should guess the posterior average y [Wolpert 1995a] As conventionally discussed, E(C d) does not bear any connection to the bias plus variance formula. However there is a mid way point between Bayesian analysis and the kind of analysis that results in the bias plus variance formula. In this middle approach, rather than fix f as in ....

[Article contains additional citation context not shown here]

Wolpert, D. (1995). "The relationship between PAC, the statistical physics framework, the Bayesian framework, and the VC framework". In The Mathematics of Generalization, D. Wolpert (Ed.), Addison-Wesley.

.... with finding the P (g j d) that optimizes P (c j d) and sampling theory statistics with evaluating P (c j t; m) m being the data set size) It is only with the extended Bayesian framework that one can consider both at once, and thereby investigate the subtle connections between the two [7]. The implicit view in this extended framework is that inference is a 2 person game pitting you, the statistician, against the data generating mechanism, aka the universe. Your opponent draws truths t at random, according to P (t) and then randomly produces a data set from t, according to P (d j ....

D.H. Wolpert, "The Relationship Between PAC, the Statistical Physics framework, the Bayesian framework, and the VC framework", in The Mathematics of Generalization, D.H. Wolpert (Ed.), Addison Wesley, 1994.

No context found.

Wolpert, D.H., (1994b). "The relationship between PAC, the Statistical Physics framework, the Bayesian framework and the VC framework" in Wolpert (1994a).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC