Learning Machines: Foundations of Trainable Pattern-classifying Systems |
Kirjan sisältä
Tulokset 1 - 3 kokonaismäärästä 14
Sivu 68
sponding region in weight space is called the solution region . It is a convex region containing all of the solution weight points W satisfying inequality ( 4.3 ) . These ideas are illustrated in Fig . 4.1 for a two - dimensional weight ...
sponding region in weight space is called the solution region . It is a convex region containing all of the solution weight points W satisfying inequality ( 4.3 ) . These ideas are illustrated in Fig . 4.1 for a two - dimensional weight ...
Sivu 84
Certainly k can be no larger than km , which is a solution to the equation kmM = km ? a2 W 2 or km = MW12 a ? ( 5.21 ) Therefore , we have proved ( for W , = 0 ) that the fixed - increment error - correction procedure must terminate ...
Certainly k can be no larger than km , which is a solution to the equation kmM = km ? a2 W 2 or km = MW12 a ? ( 5.21 ) Therefore , we have proved ( for W , = 0 ) that the fixed - increment error - correction procedure must terminate ...
Sivu 85
5.4 Proof 2 The following proof of Theorem 5.1 results from a simple geometric argument revealing that it is impossible to apply the fixed - increment error - correction procedure and remain forever outside the region of solution ...
5.4 Proof 2 The following proof of Theorem 5.1 results from a simple geometric argument revealing that it is impossible to apply the fixed - increment error - correction procedure and remain forever outside the region of solution ...
Mitä ihmiset sanovat - Kirjoita arvostelu
Yhtään arvostelua ei löytynyt.
Muita painoksia - Näytä kaikki
Yleiset termit ja lausekkeet
adjusted apply assume bank belonging to category called changes Chapter cluster committee components consider consists contains correction corresponding covariance decision surfaces define denote density depends derivation described Development discriminant functions discussed distance distribution element equal error-correction estimates example exists expression FIGURE fixed given implemented important initial layered machine linear dichotomies linear machine linearly separable matrix measurements negative networks normal Note optimum origin parameters partition pattern classifier pattern hyperplane pattern space pattern vector piecewise linear plane points positive presented probability problem proof properties proved PWL machine quadric regions respect response rule sample mean selection separable shown side solution space Stanford step Suppose theorem theory threshold training methods training procedure training sequence training subsets transformation values weight vectors zero
Viitteet tähän teokseen
A Probabilistic Theory of Pattern Recognition Luc Devroye,László Györfi,Gabor Lugosi Rajoitettu esikatselu - 1997 |