DOCSLIB.ORG

Total Jensen Divergences: Definition, Properties and K-Means++ Clustering

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Total Jensen Divergences: Definition, Properties and K-Means++ Clustering Total Jensen divergences: Definition, Properties and k-Means++ Clustering Frank Nielsen1 Richard Nock2 www.informationgeometry.org 1Sony Computer Science Laboratories, Inc. 2UAG-CEREGMIA September 2013 c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 1/19 Divergences: Distortion measures F a smooth convex function, the generator. ◮ Skew Jensen divergences: ′ Jα(p : q) = αF (p) + (1 − α)F (q) − F (αp + (1 − α)q), = (F (p)F (q))α − F ((pq)α), where (pq)γ = γp + (1 − γ)q = q + γ(p − q) and (F (p)F (q))γ = γF (p)+(1−γ)F (q)= F (q)+γ(F (p)−F (q)). ◮ Bregman divergences: B(p : q)= F (p) − F (q) − p − q, ∇F (q), lim Jα(p : q) = B(p : q), α→0 lim Jα(p : q) = B(q : p). α→1 ◮ Statistical Bhattacharrya divergence: Bhat α 1−αd ′ (p1 : p2)= − log p1(x) p2(x) ν(x)= Jα(θ1 : θ2) Z for exponential families [5]. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 2/19 Geometrically designed divergences Plot of the convex generator F . F :(x, F (x)) (q, F (q)) J(p, q) (p, F (p)) tB(p : q) B(p : q) p+q q 2 p c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 3/19 Total Bregman divergences Conformal divergence, conformal factor ρ: D′(p : q)= ρ(p, q)D(p : q) plays the rˆole of “regularizer” [8] Invariance by rotation of the axes of the design space B(p : q) tB(p : q) = = ρB (q)B(p : q), 1+ ∇F (q), ∇F (q) 1 ρB (q) = p . 1+ ∇F (q), ∇F (q) Total squared Euclideanp divergence: 1 p − q, p − q tE(p, q)= . 2 1+ q, q p c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 4/19 Total Jensen divergences 1 tB(p : q) = ρB (q)B(p : q), ρB (q)= s1+ ∇F (q), ∇F (q) tJ 1 α(p : q) = ρJ (p, q)Jα(p : q), ρJ (p, q)= 2 v1+ (F (p)−F (q)) u p−q,p−q u t Jensen-Shannon divergence, square root is a metric [2]: d d 1 2pi 1 2qi JS(p, q) = pi log + qi log 2 pi + qi 2 pi + qi Xi=1 Xi=1 Lemma The square root of the total Jensen-Shannon divergence is not a metric. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 5/19 Total Jensen divergence: Illustration F p F q ′ ′ F p ( ( ) ( ))α ′ (F (p )F (q )) ( ) F (p ) α (F (p)F (q))β ′ ′ ′ (F (p )F (q ))β Jα(p : q) ′ ′ ′ Jα(p : q ) ′ F (q) ′ tJα(p : q) ′ ′ ′ F (q ) tJα(p : q ) F ((pq)α) ′ ′ F ((p q )α) ′ O p q q (pq)α p′ (p′q′) O α c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 6/19 Total Jensen divergence: Illustration α on graph plot, β on interpolated segment Two kinds of total Jensen divergences (but one always yields closed-form) β > 1 β ∈ [0, 1] β < 0 β > 1 β ∈ [0, 1] β < 0 (F (p)F (q))β (F (p)F (q))β F ((pq)α) F ((pq)α) p q p q c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 7/19 Total Jensen divergences/Total Bregman divergences Total Jensen is not a generalization of total Bregman. limit cases α ∈ {0, 1}, we have: lim tJα(p : q) = ρJ (p, q)B(p : q) = ρB (q)B(p : q), α→0 lim tJα(p : q) = ρJ (p, q)B(q : p) = ρB (p)B(q : p), α→1 since ρJ (p, q) = ρB (q). Squared chord slope index in ρJ : ∆2 ∆⊤∇F (ǫ)∆⊤∇F (ǫ) s2 = F = = ∇F (ǫ), ∇F (ǫ) = ∇F (ǫ)2. ∆2 ∆⊤∆ c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 8/19 Conformal factor from mean value theorem When p ≃ q, ρJ (p, q) ≃ ρB (q), and the total Jensen divergence tends to the total Bregman divergence for any value of α. 1 ρJ (p, q)= = ρB (ǫ), 1+ ∇F (ǫ), ∇F (ǫ) for ǫ ∈ [p, q]. p For univariate generators, explicitly the value of ǫ: ∆ ∆ ǫ = ∇F −1 F = ∇F ∗ F , ∆ ∆ where F ∗ is the Legendre convex conjugate [5]. Stolarsky mean [7]: tJα(p : q)= ρB (ǫ)J(p : q) c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 9/19 Centroids and statistical robustness Centroids (barycenters) are minimizers of average (weighted) divergences: n L(x; w) = wi × tJα(pi : x), Xi=1 cα = arg min L(x; w), x∈X ◮ Is it unique? ◮ Is it robust to outliers [3]? Iterative convex-concave procedure (CCCP) [5] c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 10/19 Robustness of Jensen centroids (univariate generator) Theorem The Jensen centroid is robust for a strictly convex and smooth ′ p+y generator f if |f ( 2 )| is bounded on the domain X for any prescribed p. ◮ Jensen-Shannon: X = R+, f (x)= x log x − x ,f ′(x) = log(x), f ′′(x) = 1/x. ′ p+y p+y |f ( 2 )| = | log 2 | is unbounded when y → +∞. JS centroid is not robust ◮ Jensen-Burg: X = R+, f (x)= − log x, f ′(x)= −1/x, ′′ 1 f (x)= x2 ′ p+y 2 |f ( 2 )| = | p+y | is always bounded for y ∈ (0, +∞). 1 2 z(y) = 2p2 − p p + y When y →∞, we have |z(y)|→ 2p < ∞. JB centroid is robust. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 11/19 Clustering: No closed-form centroid, no cry! k-means++ [1] picks up randomly seeds, no centroid calculation. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 12/19 Divergence-based k-means++ Theorem Suppose there exist some U and V such that, ∀x, y, z: tJα(x : z) ≤ U(tJα(x : y)+ tJα(y : z)) , (triangular inequality) tJα(x : z) ≤ V tJα(z : x) , (symmetric inequality) Then the average potential of total Jensen seeding with k clusters satisfies 2 E[tJα] ≤ 2U (1 + V )(2 + log k)tJopt,α, where tJopt,α is the minimal total Jensen potential achieved by a clustering in k clusters. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 13/19 Divergence-based k-means++: Two assumptions H H: ◮ First, the maximal condition number of the Hessian of F , that is, the ratio between the maximal and minimal eigenvalue (> 0) of the Hessian of F , is upperbounded by K1. ◮ Second, we assume the Lipschitz condition on F that 2 ∆F /∆, ∆≤ K2, for some K2 > 0. Lemma Assume 0 <α< 1. Then, under assumption H, for any p, q, r ∈ S, there exists ǫ> 0 such that: 2(1 + K )K 2 1 1 tJ (p : r) ≤ 2 1 tJ (p : q)+ tJ (q : r) . α ǫ 1 − α α α α c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 14/19 Divergence-based k-means++ Corollary The total skew Jensen divergence satisfies the following triangular inequality: 2(1 + K )K 2 tJ (p : r) ≤ 2 1 (tJ (p : q)+ tJ (q : r)) . α ǫα(1 − α) α α 2(1 + K )K 2 U = 2 1 ǫ Lemma 2 Symmetric inequality condition holds for V = K1 (1 + K2)/ǫ, for some 0 <ǫ< 1. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 15/19 Total Jensen divergences: Recap Total Jensen divergence = conformal divergence with non-separable double-sided conformal factor. ◮ Invariant to axis rotation of “design space“ ◮ Equivalent to total Bregman divergences [8, 4] only when p ≃ q ◮ Square root of total Jensen-Shannon divergence is not a metric (square root of total JS is a metric). ◮ Jensen centroids are not always robust (e.g., Jensen-Shannon centroid) ◮ Total Jensen k-means++ do not require centroid computations and guaranteed approximation Interest of conformal divergences in SVM [9] (double-sided separable), in information geometry [6] (flattening). c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 16/19 Thank you. @article{totalJensen-arXiv1309.7109 , author="Frank Nielsen and Richard Nock", title="Total {J}ensen divergences: {D}efinition, Properties and $k$-Means++ Clustering", year="2013", eprint="arXiv/1309.7109" } www.informationgeometry.org c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 17/19 Bibliographic references I David Arthur and Sergei Vassilvitskii. k-means++: the advantages of careful seeding. In Proceedings of the eighteenth annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 1027–1035. Society for Industrial and Applied Mathematics, 2007. Bent Fuglede and Flemming Topsoe. Jensen-Shannon divergence and Hilbert space embedding. In IEEE International Symposium on Information Theory, pages 31–31, 2004. F. R. Hampel, P. J. Rousseeuw, E. Ronchetti, and W. A. Stahel. Robust Statistics: The Approach Based on Influence Functions. Wiley Series in Probability and Mathematical Statistics, 1986. Meizhu Liu, Baba C. Vemuri, Shun-ichi Amari, and Frank Nielsen. Shape retrieval using hierarchical total Bregman soft clustering. Transactions on Pattern Analysis and Machine Intelligence, 34(12):2407–2419, 2012. Frank Nielsen and Sylvain Boltz. The Burbea-Rao and Bhattacharyya centroids. IEEE Transactions on Information Theory, 57(8):5455–5466, August 2011. Atsumi Ohara, Hiroshi Matsuzoe, and Shun-ichi Amari. A dually flat structure on the space of escort distributions. Journal of Physics: Conference Series, 201(1):012012, 2010. c 2013 Frank Nielsen, Sony Computer Science Laboratories, Inc. 18/19 Bibliographic references II Kenneth B Stolarsky. Generalizations of the logarithmic mean. Mathematics Magazine, 48(2):87–92, 1975. Baba Vemuri, Meizhu Liu, Shun-ichi Amari, and Frank Nielsen. Total Bregman divergence and its applications to DTI analysis. IEEE Transactions on Medical Imaging, pages 475–483, 2011. Si Wu and Shun-ichi Amari. Conformal transformation of kernel functions a data dependent way to improve support vector machine classifiers.
Load more

Recommended publications
  • AMS / MAA CLASSROOM RESOURCE MATERIALS VOL 35 I I “Master” — 2011/5/9 — 10:51 — Page I — #1 I I
    AMS / MAA CLASSROOM RESOURCE MATERIALS VOL 35 i i “master” — 2011/5/9 — 10:51 — page i — #1 i i 10.1090/clrm/035 Excursions in Classical Analysis Pathways to Advanced Problem Solving and Undergraduate Research i i i i i i “master” — 2011/5/9 — 10:51 — page ii — #2 i i Chapter 11 (Generating Functions for Powers of Fibonacci Numbers) is a reworked version of material from my same name paper in the International Journalof Mathematical Education in Science and Tech- nology, Vol. 38:4 (2007) pp. 531–537. www.informaworld.com Chapter 12 (Identities for the Fibonacci Powers) is a reworked version of material from my same name paper in the International Journal of Mathematical Education in Science and Technology, Vol. 39:4 (2008) pp. 534–541. www.informaworld.com Chapter 13 (Bernoulli Numbers via Determinants)is a reworked ver- sion of material from my same name paper in the International Jour- nal of Mathematical Education in Science and Technology, Vol. 34:2 (2003) pp. 291–297. www.informaworld.com Chapter 19 (Parametric Differentiation and Integration) is a reworked version of material from my same name paper in the Interna- tional Journal of Mathematical Education in Science and Technology, Vol. 40:4 (2009) pp. 559–570. www.informaworld.com c 2010 by the Mathematical Association of America, Inc. Library of Congress Catalog Card Number 2010924991 Print ISBN 978-0-88385-768-7 Electronic ISBN 978-0-88385-935-3 Printed in the United States of America Current Printing (last digit): 10987654321 i i i i i i “master” — 2011/5/9 — 10:51 — page iii — #3 i i Excursions in Classical Analysis Pathways to Advanced Problem Solving and Undergraduate Research Hongwei Chen Christopher Newport University Published and Distributed by The Mathematical Association of America i i i i i i “master” — 2011/5/9 — 10:51 — page iv — #4 i i Committee on Books Frank Farris, Chair Classroom Resource Materials Editorial Board Gerald M.
    [Show full text]
  • Sharp Inequalities Between Hölder and Stolarsky Means of Two Positive
    Aust. J. Math. Anal. Appl. Vol. 18 (2021), No. 1, Art. 8, 42 pp. AJMAA SHARP INEQUALITIES BETWEEN HÖLDER AND STOLARSKY MEANS OF TWO POSITIVE NUMBERS MARGARITA BUSTOS GONZALEZ AND AUREL IULIAN STAN Received 29 September, 2019; accepted 15 December, 2020; published 12 February, 2021. THE UNIVERSITY OF IOWA,DEPARTMENT OF MATHEMATICS, 14 MACLEAN HALL,IOWA CITY,IOWA, USA. [email protected] THE OHIO STATE UNIVERSITY AT MARION,DEPARTMENT OF MATHEMATICS, 1465 MOUNT VERNON AVENUE,MARION,OHIO, USA. [email protected] ABSTRACT. Given any index of the Stolarsky means, we find the greatest and least indexes of the Hölder means, such that for any two positive numbers, the Stolarsky mean with the given index is bounded from below and above by the Hölder means with those indexes, of the two positive numbers. Finally, we present a geometric application of this inequality involving the Fermat-Torricelli point of a triangle. Key words and phrases: Hölder means; Stolarsky means; Monotone functions; Jensen inequality; Hermite–Hadamard in- equality. 2010 Mathematics Subject Classification. Primary 26D15. Secondary 26D20. ISSN (electronic): 1449-5910 © 2021 Austral Internet Publishing. All rights reserved. This research was started during the Sampling Advanced Mathematics for Minority Students (SAMMS) Program, organized by the Depart- ment of Mathematics of The Ohio State University, during the summer of the year 2018. The authors would like to thank the SAMMS Program for kindly supporting this research. 2 M. BUSTOS GONZALEZ AND A. I. STAN 1. INTRODUCTION The Hölder and Stolarsky means of two positive numbers a and b, with a < b, are obtained by taking a probability measure µ, whose support contains the set fa, bg and is contained in the interval [a, b], integrating the function x 7! xp, for some p 2 [−∞, 1], with respect to that probability measure, and then taking the 1=p power of that integral.
    [Show full text]
  • Stolarsky Means and Hadamard's Inequality
    JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 220, 99]109Ž. 1998 ARTICLE NO. AY975822 Stolarsky Means and Hadamard's Inequality C. E. M. Pearce Department of Applied Mathematics, The Uni¨ersity of Adelaide, Adelaide, SA 5005, Australia View metadata, citation and similar papers at core.ac.uk brought to you by CORE and provided by Elsevier - Publisher Connector J. PecaricÏ and V.Ï Simic  Faculty of Textile Technology, Uni¨ersity of Zagreb, Pierottije¨a 6, 11000, Zagreb, Croatia Submitted by A. M. Fink Received March 25, 1997 A generalization is given of the extension of Hadamard's inequality to r-convex functions. A corresponding generalization of the Fink]Mond]PecaricÏ inequalities for r-convex functions is established. Q 1998 Academic Press 1. INTRODUCTION One of the most fundamental inequalities for convex functions is that associated with the name of Hadamard. This states that if f: wxa, b ª R is convex, then 1 b faŽ.qfb Ž. HftdtŽ. F . bya a 2 Hadamard's inequality has recently been extended in two quite different ways. Recall that the integral power mean Mp of a positive function f on 99 0022-247Xr98 $25.00 Copyright Q 1998 by Academic Press All rights of reproduction in any form reserved. 100 PEARCE, PECARICÏÂ, AND ÏÂSIMIC wxa,bis a functional given by 1 1rp ¡ b p HftŽ. dt , p/0, bya a MfpŽ.s~ Ž.1.1 1 b expH ln ftdtŽ. , ps0. ¢ bya a Further, the extended logarithmic mean Lp of two positive numbers a, b is given for a s b by LapŽ.,asaand for a / b by p 1 p 1 1rp ¡ b q y a q , p / y1,0, Ž.Ž.p q 1 b y a bya LapŽ.,bs~ , psy1, ln b y ln a 1Ž.b1 1bbry , ps0.
    [Show full text]
  • An Exposition on Means Mabrouck K
    Louisiana State University LSU Digital Commons LSU Master's Theses Graduate School 2004 Which mean do you mean?: an exposition on means Mabrouck K. Faradj Louisiana State University and Agricultural and Mechanical College, [email protected] Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_theses Part of the Applied Mathematics Commons Recommended Citation Faradj, Mabrouck K., "Which mean do you mean?: an exposition on means" (2004). LSU Master's Theses. 1852. https://digitalcommons.lsu.edu/gradschool_theses/1852 This Thesis is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion in LSU Master's Theses by an authorized graduate school editor of LSU Digital Commons. For more information, please contact [email protected]. WHICH MEAN DO YOU MEAN? AN EXPOSITION ON MEANS A Thesis Submitted to the Graduate Faculty of the Louisiana State University and Agricultural and Mechanical College in partial fulfillment of the requirements for the degree of Master of Science in The Department of Mathematics by Mabrouck K. Faradj B.S., L.S.U., 1986 M.P.A., L.S.U., 1997 August, 2004 Acknowledgments This work was motivated by an unpublished paper written by Dr. Madden in 2000. This thesis would not be possible without contributions from many people. To every one who contributed to this project, my deepest gratitude. It is a pleasure to give special thanks to Professor James J. Madden for helping me complete this work. This thesis is dedicated to my wife Marianna for sacrificing so much of her self so that I may realize my dreams.
    [Show full text]
  • Janusz Matkowski GENERALIZATIONS OF
    DEMONSTRATIO MATHEMATICA Vol. XLIII No 4 2010 Janusz Matkowski GENERALIZATIONS OF LAGRANGE AND CAUCHY MEAN-VALUE THEOREMS Abstract. Some generalizations of the Lagrange Mean-Value Theorem and Cauchy Mean-Value Theorem are proved and the extensions of the corresponding classes of means are presented. 1. Introduction Recall that a function M : I2 —> I is called a mean in a nontrivial interval / C R I if it is internal, that is if min(x,y) < M(x,y) < max(i,y) for all x,y € I. The mean M is called strict if these inequalities are strict for all x,y € /, x ^ y, and symmetric if M(x,y) = M(y,x) for all x,y € I. The Lagrange Mean-Value Theorem can be formulated in the following way. If a function f : I —> R is differentiable, then there exists a strict symmetric mean L : I2 —> I such that, for all x,y € /,x ^ y, (i) Mzx-y M = meii)). If f is one-to-one then, obviously, L^ := L is uniquely determined and is called a Lagrange mean generated by /. Note that formula (1) can be written in the form x x iim-H -¥), n -¥)-m\ fl(T( „ cT , 21 + £±g-t, )=/(L(^))' x,y€l,x^y, \ x 2 2 y ' 2000 Mathematics Subject Classification: Primary 26A24. Key words and phrases: mean, mean-value theorem, Lagrange theorem, Cauchy the- orem, generalization, Lagrange mean, Cauchy mean, logrithmic mean, Stolarsky mean. 766 J. Matkowski or, setting A(x,y) := in the form f(x)-f(A(x,y)) f(A(x,y)) - f(y)\ (2) A = f'(L(x,y)), x — A(x, y) ' A(x, y) — y ) x,y€ I, x^y, which shows a relationship of the mean-value theorem and the arithmetic mean.
    [Show full text]
  • A Note on Stolarsky, Arithmetic and Logarithmic Means
    A note on Stolarsky, arithmetic and logarithmic means Vania Mascioni Department of Mathematical Sciences Ball State University Muncie, IN 47306-0490, USA [email protected] Subject Classifications: Primary: 26D20, Secondary: 26E60 Keywords: Stolarsky mean, arithmetic mean, logarithmic mean Abstract We present a way to study differences of some Stolarsky means as a way to discover new inequalities, or place known inequalities in a wider context. In particular, as an application we prove a very sharp upper bound for the difference between the arithmetic and the logarithmic means of two positive numbers. 1 Introduction and main results Alomari, Darus and Dragomir [1], as an application of Hadamard-Hermite type inequalities, give some estimates for the difference between some generalized logarithmic means and power means (see Propositions 1,2, and 3 in [1]). For example, they prove bn+1 − an+1 an + bn n(n − 1) − ≤ (b − a)2 maxfjajn−2; jbjn−2g; (n + 1)(b − a) 2 12 where a < b are real and n ≥ 2 is an integer. This inequality is very sharp, though the authors do not specify if constants such as 12 are best possible. In our Theorem 2 below (see part (c) in reference to the example we just quoted) we extend these inequalities to the full range of powers p 2 R, thus showing interesting reversals and changes that occur when p falls in certain intervals, besides putting this kind of result in a broader, perhaps more natural context (our proofs also show that the inequalities given here are optimal within their natural category). The broader context we suggest is the one provided by the large family of Stolarsky means (or difference means, as they are sometimes called | see 1 below for the definitions).
    [Show full text]
  • The Log-Convexity of Another Class of One-Parameter Means and Its Applications
    Bull. Korean Math. Soc. 49 (2012), No. 1, pp. 33{47 http://dx.doi.org/10.4134/BKMS.2012.49.1.033 THE LOG-CONVEXITY OF ANOTHER CLASS OF ONE-PARAMETER MEANS AND ITS APPLICATIONS Zhen-Hang Yang Abstract. In this paper, the log-convexity of another class one-para- meter mean is investigated. As applications, some new upper and lower bounds of logarithmic mean, new estimations for identric mean and new inequalities for power-exponential mean and exponential-geometric mean are first given. 1. Introduction and main results Let p; q 2 R and a; b 2 R+− the positive semi-axis. For a =6 b the Stolarsky mean is defined as 8 ( )1=(p−q) > q ap − bp > ; pq(p − q) =6 0; > q − q > (p a b ) > 1=p > 1 ap − bp > ; p =6 0; q = 0; < p ln a − ln b ( )1=q (1.1) Sp;q(a; b) = > 1 aq − bq > ; q =6 0; p = 0; > q ln a − ln b > ( ) > ap ln a − bp ln b 1 > exp − ; p = q =6 0; > p p :> p a − b p ab; p = q = 0; and Sp;q(a; a) = a (see [33]). Another two-parameter family of means was introduced by C. Gini in [13]. That is defined as 8 ( )1=(p−q) > ap + bp < ; p =6 q; aq(+ bq ) (1.2) Gp;q(a; b) = > ap ln a + bp ln b : exp ; p = q: ap + bp Received December 11, 2009; Revised July 29, 2011. 2010 Mathematics Subject Classification. Primary 26A48, 26A51; Secondary 26D07, 26E60. Key words and phrases.
    [Show full text]
  • Stolarsky Means of Several Variables
    Journal of Inequalities in Pure and Applied Mathematics STOLARSKY MEANS OF SEVERAL VARIABLES volume 6, issue 2, article 30, EDWARD NEUMAN 2005. Department of Mathematics Received 19 October, 2004; Southern Illinois University accepted 24 February, 2005. Carbondale, IL 62901-4408, USA Communicated by: Zs. Páles EMail: [email protected] URL: http://www.math.siu.edu/neuman/personal.html Abstract Contents JJ II J I Home Page Go Back Close c 2000 Victoria University ISSN (electronic): 1443-5756 Quit 201-04 Abstract A generalization of the Stolarsky means to the case of several variables is pre- sented. The new means are derived from the logarithmic mean of several vari- ables studied in [9]. Basic properties and inequalities involving means under discussion are included. Limit theorems for these means with the underlying measure being the Dirichlet measure are established. Stolarsky Means of Several Variables 2000 Mathematics Subject Classification: 33C70, 26D20 Key words: Stolarsky means, Dresher means, Dirichlet averages, Totally positive Edward Neuman functions, Inequalities The author is indebted to a referee for several constructive comments on the first draft of this paper. Title Page Contents Contents JJ II 1 Introduction and Notation ............................. 3 J I 2 Elementary Properties of Er,s(µ; X) ..................... 7 Go Back 3 The Mean Er,s(b; X) .................................. 14 r−s Close A Appendix: Total Positivity of Er,s (x, y) ................. 19 References Quit Page 2 of 22 J. Ineq. Pure and Appl. Math. 6(2) Art. 30, 2005 http://jipam.vu.edu.au 1. Introduction and Notation In 1975 K.B. Stolarsky [16] introduced a two-parameter family of bivariate means named in mathematical literature as the Stolarsky means.
    [Show full text]
  • Stolarsky Means of Several Variables
    Journal of Inequalities in Pure and Applied Mathematics http://jipam.vu.edu.au/ Volume 6, Issue 2, Article 30, 2005 STOLARSKY MEANS OF SEVERAL VARIABLES EDWARD NEUMAN DEPARTMENT OF MATHEMATICS SOUTHERN ILLINOIS UNIVERSITY CARBONDALE, IL 62901-4408, USA [email protected] URL: http://www.math.siu.edu/neuman/personal.html Received 19 October, 2004; accepted 24 February, 2005 Communicated by Zs. Páles ABSTRACT. A generalization of the Stolarsky means to the case of several variables is pre- sented. The new means are derived from the logarithmic mean of several variables studied in [9]. Basic properties and inequalities involving means under discussion are included. Limit the- orems for these means with the underlying measure being the Dirichlet measure are established. Key words and phrases: Stolarsky means, Dresher means, Dirichlet averages, Totally positive functions, Inequalities. 2000 Mathematics Subject Classification. 33C70, 26D20. 1. INTRODUCTION AND NOTATION In 1975 K.B. Stolarsky [16] introduced a two-parameter family of bivariate means named in mathematical literature as the Stolarsky means. Some authors call these means the extended means (see, e.g., [6, 7]) or the difference means (see [10]). For r, s ∈ R and two positive numbers x and y (x 6= y) they are defined as follows [16] 1 s xr − yr r−s , rs(r − s) 6= 0; s s r x − y 1 xr ln x − yr ln y exp − + , r = s 6= 0; r xr − yr (1.1) Er,s(x, y) = 1 xr − yr r , r 6= 0, s = 0; r(ln x − ln y) √ xy, r = s = 0.
    [Show full text]
  • Generalized Least-Squares Regressions IV: Theory and Classification Using Generalized Means
    City University of New York (CUNY) CUNY Academic Works Publications and Research Kingsborough Community College 2014 Generalized Least-Squares Regressions IV: Theory and Classification Using Generalized Means Nataniel Greene CUNY Kingsborough Community College How does access to this work benefit ou?y Let us know! More information about this work at: https://academicworks.cuny.edu/kb_pubs/77 Discover additional works at: https://academicworks.cuny.edu This work is made publicly available by the City University of New York (CUNY). Contact: [email protected] Mathematics and Computers in Science and Industry Generalized Least-Squares Regressions IV: Theory and Classication Using Generalized Means Nataniel Greene Department of Mathematics and Computer Science Kingsborough Community College, CUNY 2001 Oriental Boulevard, Brooklyn, NY 11235, USA Email: [email protected] Abstract— The theory of generalized least-squares is reformu- arithmetic mean of the square deviations in x and y and was lated here using the notion of generalized means. The generalized called Pythagorean regression previously. Here, logarithmic, least-squares problem seeks a line which minimizes the average Heronian, centroidal, identric, Lorentz, and root mean square generalized mean of the square deviations in x and y. The notion of a generalized mean is equivalent to the generating regressions are described for the rst time. Ordinary least- function concept of the previous papers but allows for a more squares regression is shown here to be equivalent to minimum robust understanding and has an already existing literature. or maximum mean regression. Regressions based on weighted Generalized means are applied to the task of constructing arithmetic means of order and weighted geometric means more examples, simplifying the theory, and further classifying generalized least-squares regressions.
    [Show full text]
  • Total Jensen Divergences: Definition, Properties and K-Means++ Clustering
    Total Jensen divergences: Definition, Properties1 and k-Means++ Clustering Frank Nielsen, Senior Member, IEEE Sony Computer Science Laboratories, Inc. 3-14-13 Higashi Gotanda, 141-0022 Shinagawa-ku Tokyo, Japan [email protected] www.informationgeometry.org Richard Nock, Non-member CEREGMIA University of Antilles-Guyane Martinique, France e-mail: [email protected] Abstract We present a novel class of divergences induced by a smooth convex function called total Jensen divergences. Those total Jensen divergences are invariant by construction to rotations, a feature yielding arXiv:1309.7109v1 [cs.IT] 27 Sep 2013 regularization of ordinary Jensen divergences by a conformal factor. We analyze the relationships be- tween this novel class of total Jensen divergences and the recently introduced total Bregman divergences. We then proceed by defining the total Jensen centroids as average distortion minimizers, and study their robustness performance to outliers. Finally, we prove that the k-means++ initialization that bypasses explicit centroid computations is good enough in practice to guarantee probabilistically a constant approximation factor to the optimal k-means clustering. Index Terms September 30, 2013 DRAFT Total Bregman divergence, skew Jensen divergence, Jensen-Shannon divergence, Burbea-Rao diver- gence, centroid robustness, Stolarsky mean, clustering, k-means++. I. INTRODUCTION:GEOMETRICALLY DESIGNED DIVERGENCES FROM CONVEX FUNCTIONS A divergence D(p : q) is a smooth distortion measure that quantifies the dissimilarity between any two data points p and q (with D(p : q) = 0 iff. p = q). A statistical divergence is a divergence between probability (or positive) measures. There is a rich literature of divergences [1] that have been proposed and used depending on their axiomatic characterizations or their empirical performance.
    [Show full text]
  • Entropic Means
    JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 139, 537-551 (1989) Entropic Means AHARON BEN-TAL* Faculty of Industrial Engineering and Management, Technion, Israel Institute of Technology, Ha$a. Israel, and Department of Industrial and Operations Engineering, The University of Michigan, Ann Arbor, Michigan 48109 ABRAHAM CHARNES Center for Cybernetic Studies, The University of Texas, Ausrin, Texas 78712 AND MARC TEBOULLE~ Department of Mathematics, University of Maryland Baltimore County, Catonsville, Maryland 21228 Submitted by J. L. Brenner Received April 6, 1987 We show how to generate means as optimal solutions of a minimization problem whose objective function is an entropy-like functional. Accordingly, the resulting mean is called entropic mean. It is shown that the entropic mean satisfies the basic properties of a weighted homogeneous mean (in the sense of J. L. Brenner and B. C. Carbon (J. Math. Anal. Appl. 123 (1987), 265-280)) and that all the classical means as well as many others are special cases of entropic means, Comparison theorems are then proved and used to derive inequalities between various means. Generaliza- tions to entropic means of random variables are considered. An extremal principle for generating the Hardy-Littlewood-Polya generalized mean is also derived. Finally, it is shown that an asymptotic formula originally discovered by L. Hoehn and I. Niven (Mafh. Mag. 58 (1958), 151-156) for power means, and later extended by J. L. Brenner and B. C. Carlson (J. Math. Anal. Appl. 123(1987), 265-280) is valid for entropic means. c 1989 Academic Press, lnc * This work was supported in part by the NSF Grant ECS-860-4354.
    [Show full text]