Learning Machines: Foundations of Trainable Pattern-classifying SystemsMcGraw-Hill, 1965 - 137 sivua |
Kirjan sisältä
Tulokset 1 - 3 kokonaismäärästä 38
Sivu 83
... W , such that Y. W > 0 for all Y in y ' . For a fixed solution vector W let min Y. WA a Yey ' ( 5.11 ) where a > 0. Taking the dot product of the solution vector W with both sides of Eq . ( 5 · 10 ) yields Ŵk + 1 • 1 W = Ŷ1 • W + Ŷ2 • W ...
... W , such that Y. W > 0 for all Y in y ' . For a fixed solution vector W let min Y. WA a Yey ' ( 5.11 ) where a > 0. Taking the dot product of the solution vector W with both sides of Eq . ( 5 · 10 ) yields Ŵk + 1 • 1 W = Ŷ1 • W + Ŷ2 • W ...
Sivu 86
... W , d + 1 , effected by the kth step dk + 1 = - | W – Ŵx | 2 - W - Wx + 12 ( 5.24 ) Let Y be the kth pattern vector in the reduced training sequence St. W. /w.Y1 = ( M + b ) • = 0 " Insulated " region , W ' Origin Pattern hyperplanes ...
... W , d + 1 , effected by the kth step dk + 1 = - | W – Ŵx | 2 - W - Wx + 12 ( 5.24 ) Let Y be the kth pattern vector in the reduced training sequence St. W. /w.Y1 = ( M + b ) • = 0 " Insulated " region , W ' Origin Pattern hyperplanes ...
Sivu 92
... W ( 5.38 ) k for all W in W. We therefore say that Ŵ + 1 is pointwise closer than Ŵ1⁄2 to W. As a first step in proving the theorem , we shall show that the sequence S converges to a point P. = k For any fixed W in W let lim | W - W 1 ( W ) ...
... W ( 5.38 ) k for all W in W. We therefore say that Ŵ + 1 is pointwise closer than Ŵ1⁄2 to W. As a first step in proving the theorem , we shall show that the sequence S converges to a point P. = k For any fixed W in W let lim | W - W 1 ( W ) ...
Sisältö
TRAINABLE PATTERN CLASSIFIERS | 1 |
SOME NONPARAMETRIC TRAINING METHODS | 65 |
LAYERED MACHINES | 95 |
Tekijänoikeudet | |
1 muita osia ei näytetty
Muita painoksia - Näytä kaikki
Yleiset termit ja lausekkeet
assume belonging to category Chapter cluster committee machine committee TLUS components correction increment covariance matrix decision surfaces denote diagonal matrix dot products error-correction procedure Euclidean distance example Fix and Hodges function g(X g₁(X gi(X given Hodges method hypersphere image-space implemented initial weight vectors ith bank layer of TLUS layered machine linear dichotomies linear discriminant functions linearly separable loss function mean vector minimum-distance classifier mode-seeking networks nonparametric number of patterns p₁ parameters parametric training partition pattern hyperplane pattern points pattern space pattern vector pattern-classifying patterns belonging perceptron piecewise linear plane point sets positive probability distributions prototype pattern PWL machine quadratic form quadric function rule sample covariance matrix shown in Fig solution weight vector Stanford subsets X1 subsidiary discriminant functions Suppose terns training patterns training sequence training set training subsets transformation two-layer machine values W₁ wa+1 weight point weight space weight-vector sequence X1 and X2 zero
Viitteet tähän teokseen
A Probabilistic Theory of Pattern Recognition Luc Devroye,László Györfi,Gabor Lugosi Rajoitettu esikatselu - 1997 |