E. Nelson, The free Marko eld, J. Functional Analysis 12, 211-227 (1973)  Home/Search   Document Not in Database   Summary   Related Articles   Check

....routes to estimate the random averages, especially from above since this is the direction one needs for sample complexity bounds. We show that it is possible to bound the Rademacher and gaussian averages 25 using the empirical L 2 entropy of the class. This follows from results due to Dudley [6] and Sudakov [31] Originally, the bounds were established from gaussian processes, and later they were extended to the sub gaussian setup ( 8, 34] which includes Rademacher processes. Theorem 2.31 There are absolute constants C and c for which the following holds. For any integer n, any ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

....to all nite dimensional subspaces of L 2 . Thus, if I : n 2 (IR n ; k k K ) is the formal identity operator, then (F ) I) The following deep result provides a connection between the norm of a set and its covering numbers in L 2 . The upper bound was established by Dudley in [4] while the lower bound is due to Sudakov (see [17] A proof of both bounds may be found in [14] 7 Theorem 2.11 Let F n 2 . Then there are absolute positive constants c and C (i.e. independent of F and n) such that c sup 0 log 1 2 N( F ) F ) C Z 1 0 log 1 2 ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

.... where the subset F is obvious, we shall use the notation N( d) and in cases where the metric is clear we shall denote the covering numbers by N( F) The next theorem demonstrates the connection between the covering numbers of a set F and (F ) The upper bound was established by Dudley in [2], while the lower one is due to Sudakov (see [13] A proof of both bounds may be found in [9] Theorem 2.3 Let F n 2 . Then, there are absolute positive constants c and C such that c sup 0 p log(N( F ) F ) C Z 1 0 p log(N( F ) d (2.2) Next, we examine some basic de ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

....standard Gaussian random variables. First, note that absconv(F ) n = F= n ) Much less obvious and considerably more important is the following result, which provides an upper bound on (F= n ) in terms of the covering numbers of F in L 2 ( n ) This bound was demonstrated by Dudley [6] Theorem 2. There is an absolute constant C such that for every integer n and every F L 2 ( n ) F= n ) C Z 1 0 log 1 2 N( F= n ; L 2 ( n ) d : In a similar fashion, it is possible to de ne the Rademacher averages associated with a class F and a sample f 1 ; n g. De ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

....following connection: Theorem 2.9 There is an absolute constant C such that for every integer n and every F n 2 , CR(F ) F ) The following deep result provides a connection between the norm of a set and its covering numbers in n 2 . The upper bound was established by Dudley in [5] while the lower bound is due to Sudakov [17] A proof of both bounds may be found in [16] Theorem 2.10 Let F n 2 . Then there are absolute positive constants c and C, such that c sup 0 log 1 2 N( F; n 2 ) F ) C Z 1 0 log 1 2 N( F; n 2 ) d : 3 The ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

....use a well know geometric parameter from the local theory of Banach spaces called the norm. This parameter measures how large a given set is. It is possible to establish both upper and a lower bounds on (F) by the L 2 log covering numbers of the set F . This important fact is due to Dudley ([4]) and Sudakov ( 14] In section 3 and in the Appendix we investigate the norms of GC classes in empirical L 2 spaces. We provide an upper bound to (F) in terms of V C(F) or P (F) The norm estimates enable us to bound the number of the statistics required for suciency. Recall that a set ....

....of F [ F , then E n X i=1 g i e i 2 K = Z IR n sup f2F[F x; f 2 d n ; implying that (K) F ) The following deep result provides a connection between the norm of a set and its covering numbers in n 2 . The upper bound was established by Dudley in [4] while the lower one is due to Sudakov (see [14] A proof of both bounds may be found in [12] Theorem 3.2 Let F n 2 . Then there are absolute positive constants c and C such that c sup 0 log 1 2 (N( F ) F ) C Z 1 0 log 1 2 (N( F ) d : If F is a class of functions ....

R.M. Dudley: The sizes of compact subsets of Hilbert space and continuity of Gaussian processes, J. of Functional Analysis 1, 290-330, 1967.

....model for determines Rieszness; more specifically it leads us to conjecture that the no twisted sectors property introduced by Shapiro, Smith, and Stegenga in [18] is equivalent to Rieszness. After this work was completed Pietro Poggi Corradini established the validity of this conjecture [12]. In the final section of this paper we discuss the no twisted sectors conjecture in more detail, and point out a connection between our work and a recent result of Cowen and MacCluer [5, Corollary 19] concerning compositionoperator spectra. 2. Background. For completeness of exposition we ....

....sector. More generally, we conjecture that for an arbitrary holomorphic self map (U) satisfying (0) 0 and 0 j 0 (0)j 1) the implication (8) goes both ways: C : H 2 H 2 is Riesz if and only if oe 2 H p for all p 1. As we mentioned in the Introduction, Poggi Corradini [12] has recently established the first of these conjectures. We have just learned that he has proved the second as well [14] We conclude by noting a connection between our work and the study of composition operator spectra. In [5] Cowen and MacCluer characterize the spectrum of C given that is ....

, The Hardy class of geometric models and the essential spectral radius of composition operators, J. Functional Analysis, 143 (1997), 129-156.

No context found.

E. Nelson, The free Marko eld, J. Functional Analysis 12, 211-227 (1973)

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