Q-Learning Algorithms with Random Truncation Bounds and Applications to Effective Parallel Computing

Motivated by an important problem of load balancing in parallel computing, this paper examines a modified algorithm to enhance Q-learning methods, especially in asynchronous recursive procedures for self-adaptive load distribution at run-time. Unlike the existing projection method that utilizes a fixed region, our algorithm employs a sequence of growing truncation bounds to ensure the boundedness of the iterates. Convergence and rates of convergence of the proposed algorithm are established. This class of algorithms has broad applications in signal processing, learning, financial engineering, and other related fields. G. Yin’s research was supported in part by the National Science Foundation under Grants DMS-0603287 and DMS-0624849 and in part by the National Security Agency under Grant MSPF-068-029. C.Z. Xu’s research was supported in part by the National Science Foundation under Grants CCF-0611750, DMS-0624849, CNS-0702488, and CRI-0708232. L.Y. Wang’s research was supported in part by the National Science Foundation under Grants ECS-0329597 and DMS-0624849 and by the Michigan Economic Development Council.     

On the distribution of imaginary parts of zeros of the Riemann zeta function,II

We continue our investigation of the distribution of the fractional parts of αγ, where α is a fixed non-zero real number and γ runs over the imaginary parts of the non-trivial zeros of the Riemann zeta function. We establish some connections to Montgomery’s pair correlation function and the distribution of primes in short intervals. We also discuss analogous results for a more general L-function. The first author is supported by National Science Foundation Grant DMS-0555367. The second author is partially supported by the National Science Foundation and the American Institute of Mathematics (AIM). The third author is supported by National Science Foundation Grant DMS-0456615.     

Local Linear Convergence for Alternating and Averaged Nonconvex Projections

The idea of a finite collection of closed sets having “linearly regular intersection” at a point is crucial in variational analysis. This central theoretical condition also has striking algorithmic consequences: in the case of two sets, one of which satisfies a further regularity condition (convexity or smoothness, for example), we prove that von Neumann’s method of “alternating projections” converges locally to a point in the intersection, at a linear rate associated with a modulus of regularity. As a consequence, in the case of several arbitrary closed sets having linearly regular intersection at some point, the method of “averaged projections” converges locally at a linear rate to a point in the intersection. Inexact versions of both algorithms also converge linearly. Research of A.S. Lewis supported in part by National Science Foundation Grant DMS-0504032. Research of D.R. Luke supported in part by National Science Foundation Grant DMS-0712796.     

On the eigenvalues of matrices with given upper triangular part

We give necessary and sufficient conditions for a set of numbers to be the eigenvalues of a completion of a matrix prescribed in its upper triangular part.Partially supported by the NSF Grant DMS-8701615-02Partially supported by the NSF Grant DMS-8802836 and United States-Israel Binational Science Foundation Grant 88-00304/I.     

The Gaussian approximation for multi-color generalized Friedman’s urn model

The generalized Friedman’s urn model is a popular urn model which is widely used in many disciplines.In particular,it is extensively used in treatment allocation schemes in clinical trials.In this paper,we show that both the urn composition process and the allocation proportion process can be approximated by a multi-dimensional Gaussian process almost surely for a multi-color generalized Friedman’s urn model with both homogeneous and non-homogeneous generating matrices.The Gaussian process is a solution of ...     

Consolidation rates for two interacting systems in the plane

Summary This paper is a sequel of a paper of Cox and Griffeath “diffusive clustering in the two dimensional voter model”. We continue our study of the voter model and coalescing random walks on the two dimensional integer lattice. Some exact asymptotics concerning the rate of clustering in the former process and the coalescence rate of the latter are derived. We use these results to prove a limit law, announced in that earlier paper, concerning the size of the largest square centered at the origin which is of solid color at a large time t. Partially supported by the National Science Foundation under Grant DMS-831080 Partially supported by the National Science Foundation under Grant DMS-841317 Partially supported by the National Science Foundation under Grant DMS-830549     

On complex eigenvalues ofM andP matrices

Summary Inequalities are obtained for the complex eigenvalues of anM matrix or aP matrix which depend only on the order of the matrix.This research was supported in part by the National Science Foundation under Grant GP-20555.     

Stochastic regression and its application to hedging in finance

In this paper we investigate how to employ stochastic regression to hedge risks in finance, where the risk of a security is measured by its quadratic variation process. Mykland and Zhang used this technique to demonstrate how to reduce the risk of a given security by introducing another security. In this paper, we investigate how to further reduce the remaining unhedgable risk by adding more hedging securities. Some practical guidelines on how to choose those hedging securities in practice is also given. Jing’s research was partially supported by Hong Kong RGC (Grant Nos. HKUST6011/07P, HKUST6015/08P), and Zhang’s research was supported in part by National Natural Science Foundation of China (Grant No. 10771214)     

Entrywise relative perturbation bounds for exponentials of essentially non-negative matrices

A real square matrix is said to be essentially non-negative if all of its off-diagonal entries are non-negative. We establish entrywise relative perturbation bounds for the exponential of an essentially non-negative matrix. Our bounds are sharp and contain a condition number that is intrinsic to the exponential function. As an application, we study sensitivity of continuous-time Markov chains. J. Xue was supported by the National Science Foundation of China under grant number 10571031, the Program for New Century Excellent Talents in Universities of China and Shanghai Pujiang Program. Q. Ye was supported in part by NSF under Grant DMS-0411502.     

Simultaneously inpainting in image and transformed domains

In this paper, we focus on the restoration of images that have incomplete data in either the image domain or the transformed domain or in both. The transform used can be any orthonormal or tight frame transforms such as orthonormal wavelets, tight framelets, the discrete Fourier transform, the Gabor transform, the discrete cosine transform, and the discrete local cosine transform. We propose an iterative algorithm that can restore the incomplete data in both domains simultaneously. We prove the convergence of the algorithm and derive the optimal properties of its limit. The algorithm generalizes, unifies, and simplifies the inpainting algorithm in image domains given in Cai et al. (Appl Comput Harmon Anal 24:131–149, 2008) and the inpainting algorithms in the transformed domains given in Cai et al. (SIAM J Sci Comput 30(3):1205–1227, 2008), Chan et al. (SIAM J Sci Comput 24:1408–1432, 2003; Appl Comput Harmon Anal 17:91–115, 2004). Finally, applications of the new algorithm to super-resolution image reconstruction with different zooms are presented. R. H. Chan’s research was supported in part by HKRGC Grant 400505 and CUHK DAG 2060257. L. Shen’s research was supported by the US National Science Foundation under grant DMS-0712827. Z. Shen’s research was supported in part by Grant R-146-000-060-112 at the National University of Singapore.     

A convergence theorem of Rosen’s gradient projection method

Since Rosen’s gradient projection method was published in 1960, a rigorous convergence proof of his method has remained an open question. A convergence theorem is given in this paper. Part of this author’s work was done while he studied at the Department of Mathematics, University of California at Santa Barbara, and was supported by the National Science Foundation under Grant No. MCS83-14977. Part of this author’s work was done while he visited the Computer Science Department, University of Minnesota, Minneapolis, and was supported by the National Science Foundation under Grant No. MCS81-01214.     

Approximation of analytic functions with prescribed boundary conditions by circle-packing maps

We use recent advances in circle-packing theory to develop a constructive method for the approximation of an analytic functionF: Ω →C by circle packing maps providing we have only been given ΩF’_dΩ, and the set of critical points ofF. This extends the earlier results of Carter and Rodin and of Colin de Verdière and Mathéus, for functionsF with no critical points. The author gratefully acknowledges support of the Tennessee Science Alliance and the National Science Foundation. Research at MSRI is supported in part by Grant No. DMS-9022140.     

Proper holomorphic self-maps of smooth bounded Reinhardt domains in ?2

It is proved that every proper holomorphic self-map of a smooth bounded Reinhardt domain in ℂ² is an automorphism. The first author’s work was supported in part by the National Natural Science Foundation of China (Grant No. 10571135) and the Doctoral Program Foundation of the Ministry of Education of China (Grant No. 20050240711)     

Characterizations of spaces related to metasymplectic spaces

This research was partially supported by the National Science Foundation under Grant Number DMS-8691779.     

Superconvergence in the generalized finite element method

In this paper, we address the problem of the existence of superconvergence points of approximate solutions, obtained from the Generalized Finite Element Method (GFEM), of a Neumann elliptic boundary value problem. GFEM is a Galerkin method that uses non-polynomial shape functions, and was developed in (Babuška et al. in SIAM J Numer Anal 31, 945–981, 1994; Babuška et al. in Int J Numer Meth Eng 40, 727–758, 1997; Melenk and Babuška in Comput Methods Appl Mech Eng 139, 289–314, 1996). In particular, we show that the superconvergence points for the gradient of the approximate solution are the zeros of a system of non-linear equations; this system does not depend on the solution of the boundary value problem. For approximate solutions with second derivatives, we have also characterized the superconvergence points of the second derivatives of the approximate solution as the roots of a system of non-linear equations. We note that smooth generalized finite element approximation is easy to construct. I. Babuška’s research was partially supported by NSF Grant # DMS-0341982 and ONR Grant # N00014-99-1-0724. U. Banerjee’s research was partially supported by NSF Grant # DMS-0341899. J. E. Osborn’s research was supported by NSF Grant # DMS-0341982.     

Continuity in weak topology:higher order linear systems of ODE

We will introduce a type of Fredholm operators which are shown to have a certain con- tinuity in weak topologies.From this,we will prove that the fundamental matrix solutions of k-th, k≥2,order linear systems of ordinary differential equations are continuous in coefficient matrixes with weak topologies.Consequently,Floquet multipliers and Lyapunov exponents for periodic systems are continuous in weak topologies.Moreover,for the scalar Hill's equations,Sturm-Liouville eigenvalues, periodic and anti-periodic eigenvalues,and rotation numbers are all continuous in potentials with weak topologies.These results will lead to many interesting variational problems.     

A robust sequential quadratic programming method

The sequential quadratic programming method developed by Wilson, Han and Powell may fail if the quadratic programming subproblems become infeasible, or if the associated sequence of search directions is unbounded. This paper considers techniques which circumvent these difficulties by modifying the structure of the constraint region in the quadratic programming subproblems. Furthermore, questions concerning the occurrence of an unbounded sequence of multipliers and problem feasibility are also addressed.Work supported in part by the National Science Foundation under Grant No. DMS-8602399 and by the Air Force Office of Scientific Research under Grant No. ISSA-860080.Work supported in part by the National Science Foundation under Grant No. DMS-8602419.     

Index formulas for geometric Dirac operators in Riemannian foliations

With regards to certain Riemannian foliations we consider Kasparov pairings of leafwise and transverse Dirac operators. Relative to a pairing with a transversal class we commence by establishing an index formula for foliations with leaves of nonpositive sectional curvature. The underlying ideas are then developed in a more general setting leading to pairings of images under the Baum-Connes map in geometricK-theory with transversal classes. Several ideas implicit in the work of Connes and Hilsum-Skandalis are formulated in the context of Riemannian foliations. From these we establish the notion of a dual pairing inK-homology and a theorem of the Grothendieck-Riemann-Roch type.R. G. D. was supported by The National Science Foundation under Grant No. DMS-9304283.J. F. G. and F. W. K. were supported in part by The National Science Foundation under Grant No. DMS-9208182.F. W. K. was also supported in part by an Arnold O. Beckman Research Award from the Research Board of the University of Illinois.     

A nonconforming finite element method for a two-dimensional curl–curl and grad-div problem

A numerical method for a two-dimensional curl–curl and grad-div problem is studied in this paper. It is based on a discretization using weakly continuous P ₁ vector fields and includes two consistency terms involving the jumps of the vector fields across element boundaries. Optimal convergence rates (up to an arbitrary positive ) in both the energy norm and the L ₂ norm are established on graded meshes. The theoretical results are confirmed by numerical experiments. The work of the first author was supported in part by the National Science Foundation under Grant No. DMS-03-11790 and by the Humboldt Foundation through her Humboldt Research Award. The work of the third author was supported in part by the National Science Foundation under Grant No. DMS-06-52481.     

Sizer for smoothing splines

Smoothing splines are an attractive method for scatterplot smoothing. The SiZer approach to statistical inference is adapted to this smoothing method, named SiZerSS. This allows quick and sure inference as to “which features in the smooth are really there” as opposed to “which are due to sampling artifacts”, when using smoothing splines for data analysis. Applications of SiZerSS to mode, linearity, quadraticity and monotonicity tests are illustrated using a real data example. Some small scale simulations are presented to demonstrate that the SiZerSS and the SiZerLL (the original local linear version of SiZer) often give similar performance in exploring data structure but they can not replace each other completely. Marron’s research was supported by the Dept. of Stat. and Appl. Prob., National Univ. of Singapore, and by the National Science Foundation Grant DMS-9971649. Zhang’s research was supported by the National Univ. of Singapore Academic Research grant R-155-000-023-112. The Editor, the Associate Editor, and the referees are appreciated for their invaluable comments and suggestions that help improve the article significantly.