Abstract: A theoretical justification for the random vector version of the functional-link (RVFL) net is presented in this paper, based on a general approach to adaptive function approximation The approach consists of formulating a limit-integral representation of the function to be approximated and subsequently evaluating that integral with the Monte-Carlo method Two main results are: (1) the RVFL is a universal approximator for continuous functions on bounded finite dimensional sets, and (2) the RVFL is an efficient universal approximator with the rate of approximation error convergence to zero of order O(C//spl radic/n), where n is number of basis functions and with C independent of n Similar results are also obtained for neural nets with hidden nodes implemented as products of univariate functions or radial basis functions Some possible ways of enhancing the accuracy of multivariate function approximations are discussed >

Abstract: An analytic expression is proposed for the three-point function of the exponential fields in the Liouville field theory on a sphere. In the classical limit it coincides with what the classical Liouville theory predicts. Using this function as the structure constant of the operator algebra we construct the four-point function of the exponential fields and verify numerically that it satisfies the conformal bootstrap equations, i.e., that the operator algebra thus defined is associative. We consider also the Liouville reflection amplitude which follows explicitly from the structure constants.

Abstract: We present serial and parallel algorithms for solving a system of equations that arises from the discretization of the Hamilton-Jacobi equation associated to a trajectory optimization problem of the following type. A vehicle starts at a prespecified point x/sub o/ and follows a unit speed trajectory x(t) inside a region in /spl Rscr//sup m/ until an unspecified time T that the region is exited. A trajectory minimizing a cost function of the form /spl int//sub 0//sup T/ r(x(t))dt+q(x(T)) is sought. The discretized Hamilton-Jacobi equation corresponding to this problem is usually solved using iterative methods. Nevertheless, assuming that the function r is positive, we are able to exploit the problem structure and develop one-pass algorithms for the discretized problem. The first algorithm resembles Dijkstra's shortest path algorithm and runs in time O(n log n), where n is the number of grid points. The second algorithm uses a somewhat different discretization and borrows some ideas from a variation of Dial's shortest path algorithm (1969) that we develop here; it runs in time O(n), which is the best possible, under some fairly mild assumptions. Finally, we show that the latter algorithm can be efficiently parallelized: for two-dimensional problems and with p processors, its running time becomes O(n/p), provided that p=O(/spl radic/n/log n). >

Abstract: The success of reinforcement learning in practical problems depends on the ability to combine function approximation with temporal difference methods such as value iteration. Experiments in this area have produced mixed results; there have been both notable successes and notable disappointments. Theory has been scarce, mostly due to the difficulty of reasoning about function approximators that generalize beyond the observed data. We provide a proof of convergence for a wide class of temporal difference methods involving function approximators such as k-nearest-neighbor, and show experimentally that these methods can be useful. The proof is based on a view of function approximators as expansion or contraction mappings. In addition, we present a novel view of approximate value iteration: an approximate algorithm for one environment turns out to be an exact algorithm for a different environment.

Abstract: We provide criteria for positive definiteness of radial functions with compact support. Based on these criteria we will produce a series of positive definite and compactly supported radial functions, which will be very useful in applications. The simplest ones arecut-off polynomials, which consist of a single polynomial piece on [0, 1] and vanish on [1, ∞). More precisely, for any given dimensionn and prescribedCk smoothness, there is a function inCk(ℝn), which is a positive definite radial function with compact support and is a cut-off polynomial as a function of Euclidean distance. Another example is derived from odd-degreeB-splines.

Abstract: In this paper, it is shown that the problem of checking the solvability of a bilinear matrix inequality (BMI), is NP-hard. A matrix valued function, F(X,Y), is called bilinear if it is linear with respect to each of its arguments, and an inequality of the form, F(X,Y)>0 is called a bilinear matrix inequality. Recently, it was shown that, the static output feedback problem, fixed order controller problem, reduced order H/sup /spl infin// controller design problem, and several other control problems can be formulated as BMIs. The main result of this paper shows that the problem of checking the solvability of BMIs is NP-hard, and hence it is rather unlikely to find a polynomial time algorithm for solving general BMI problems. As an independent result, it is also shown that simultaneous stabilization with static output feedback is an NP-hard problem, namely for given n plants, the problem of checking the existence of a static gain matrix, which stabilizes all of the n plants, is NP-hard.

Abstract: We prove that the work function algorithm for the k-server problem has a competitive ratio at most 2k−1. Manasse et al. [1988] conjectured that the competitive ratio for the k-server problem is exactly k (it is trivially at least k); previously the best-known upper bound was exponential in k. Our proof involves three crucial ingredients: A quasiconvexity property of work functions, a duality lemma that uses quasiconvexity to characterize the configuration that achieve maximum increase of the work function, and a potential function that exploits the duality lemma.

Abstract: Within the framework of pac-learning, we explore the learnability of concepts from samples using the paradigm of sample compression schemes. A sample compression scheme of size k for a concept class C ⊆ 2X consists of a compression function and a reconstruction function. The compression function receives a finite sample set consistent with some concept in C and chooses a subset of k examples as the compression set. The reconstruction function forms a hypothesis on X from a compression set of k examples. For any sample set of a concept in C the compression set produced by the compression function must lead to a hypothesis consistent with the whole original sample set when it is fed to the reconstruction function. We demonstrate that the existence of a sample compression scheme of fixed-size for a class C is sufficient to ensure that the class C is pac-learnable. Previous work has shown that a class is pac-learnable if and only if the Vapnik-Chervonenkis (VC) dimension of the class is finite. In the second half of this paper we explore the relationship between sample compression schemes and the VC dimension. We define maximum and maximal classes of VC dimension d. For every maximum class of VC dimension d, there is a sample compression scheme of size d, and for sufficiently-large maximum classes there is no sample compression scheme of size less than d. We discuss briefly classes of VC dimension d that are maximal but not maximum. It is an open question whether every class of VC dimension d has a sample compression scheme of size O(d).

Abstract: "Constructive wavelet networks" are investigated as a universal tool for function approximation. The parameters of such networks are obtained via some "direct" Monte Carlo procedures. Approximation bounds are given. Typically, it is shown that such networks with one layer of "wavelons" achieve an L/sub 2/ error of order O(N/sup -(/spl rhod)/), where N is the number of nodes, d is the problem dimension and /spl rho/ is the number of summable derivatives of the approximated function. An algorithm is also proposed to estimate this approximation based on noisy input-output data observed from the function under consideration. Unlike neural network training, this estimation procedure does not rely on stochastic gradient type techniques such as the celebrated "backpropagation" and it completely avoids the problem of poor convergence or undesirable local minima. >

Abstract: We describe a machine learning method for predicting the value of a real-valued function, given the values of multiple input variables. The method induces solutions from samples in the form of ordered disjunctive normal form (DNF) decision rules. A central objective of the method and representation is the induction of compact, easily interpretable solutions. This rule-based decision model can be extended to search efficiently for similar cases prior to approximating function values. Experimental results on real-world data demonstrate that the new techniques are competitive with existing machine learning and statistical methods and can sometimes yield superior regression performance.

Abstract: Let $\hat{F}_n$ be the Kaplan-Meier estimator of a distribution function $F$ computed from randomly censored data. We show that under optimal integrability assumptions on a function $\varphi$, the Kaplan-Meier integral $\int \varphi d\hat{F}_n$, when properly standardized, is asymptotically normal.

Abstract: In the course of the studies on fuzzy regression analysis, we encountered the problem of introducing a distance between fuzzy numbers, which replaces the classical (x - y)2 on the real line. Our proposal is to compute such a function as a suitable weighted mean of the distances between the a-cuts of the fuzzy numbers. The main difficulty is concerned with the definition of the distance between intervals, since the current definitions present some disadvantages which are undesirable in our context. In this paper we describe an approach which removes such drawbacks.

Abstract: This thesis addresses the non-linear system identification problem, and in particular, investigates the use of neural networks in system identification. An overview of different possible mode! structures is given in a common framework. A nonlinear structure is described as the concatenation of a map from the observed data to the regressor, and a map from the regressor to the output space. This divides the model structure selection problem into two problems with lower complexity: that of choosing the regressor and that of choosing the non-linear map.The possible choices for the regressors consists of past inputs and outputs, and filtered versions of them. The dynamics of the mode! depends on the choice of regressor, and families of different mode! structures are suggested based on analogies to linear black-box models. State-space models are also described within this common framework by a special choice of regressor. It is shown that state-space models which have no parameters in the state update function can be viewed as an input-output mode! preceded by a pre-filter. A parameterized state update function, on the other hand, can be seen as a data driven regressor selector. The second step of the nonlinear identification is the mapping from the regressor to the output space. It is often advantageous to try some intermediate mappings between the linear and the general non-linear mapping. Such non-linear black-box mappings are discussed and motivated by considering different noise assumptions.The validation of a linear mode! should contain a test for non-linearities and it is shown that, in general, it is easy to detect non-linearities. This implies that it is not worth spending too much energy searching for optimal non-linear validation methods for a specific problem. lnstead the validation method should be chosen so that it is easy to apply. Two such methods, based on polynomials and neural nets, are suggested. Further, two validation methods, the correlation-test and the parametric F-test, are investigated. It is shown that under certain conditions these methodscoincide.Parameter estimates are usually based on criterion minimization. In connection with neural nets it has been noted that it is not always optimal to try to find the absolute minimum point of the criterion. Instead a better estimate can be obtained if the numerical search for the minimum is prematurely stopped. A forma! connection between this stopped search and regularization is given. It is shown that the numerical minimization of the criterion can be view as a regularization term which is gradually turned to zero. This closely connects to, and explains, what is called overtraining in the neural net literature.

Abstract: A low-degree test is a collection of simple, local rules for checking the proximity of an arbitrary function to a low-degree polynomial. Each rule depends on the function's values at a small number of places. If a function satisfies many rules then it is close to a low-degree polynomial. Low-degree tests play an important role in the development of probabilistically checkable proofs. In this paper we present two improvements to the efficiency of low-degree tests. Our first improvement concerns the smallest field size over which a low-degree test can work. We show how to test that a function is a degree d polynomial over prime fields of size only d+2. Our second improvement shows a better efficiency of the low-degree test of Rubinfeld and Sudan (1993) than previously known. We show concrete applications of this improvement via the notion of "locally checkable codes". This improvement translates into better tradeoffs on the size versus probe complexity of probabilistically checkable proofs than previously known. >

Abstract: This paper addresses the problem of optimizing a function over a finite or countably infinite set of alternatives, in situations where this objective function cannot be evaluated exactly, but has to be estimated or measured. A special focus is on situations where simulation is used to evaluate the objective function. We present two versions of a new iterative method for solving such discrete stochastic optimization problems. In each iteration of the proposed method, a neighbor of the “current” alternative is selected, and estimates of the objective function evaluated at the current and neighboring alternatives are compared. The alternative that has a better observed function value becomes the next current alternative. We show how one version of the proposed method can be used to solve discrete optimization problems where the objective function is evaluated using transient or steady-state simulation, and we show how the other version can be applied to solve a special class of discrete stochastic optimizati...

Abstract: A method is described for network optimization based on a multirate, circuit-switched analysis. Network loss probabilities are determined as a solution of a set of fixed point equations and the sensitivity of network performance, as a function of offered load and loss probabilities, is determined as a solution to a set of linear equations. Because the numerical complexity of solving both the fixed point equations and the sensitivity equations is of an order which renders an exact solution computationally intractable, an asymptotic approximation is applied which yields a solution to the network loss probabilities and network sensitivities. A global optimization procedure is then applied using an iterative, steepest ascent optimization procedure to yield a set of virtual path routings and capacity allocations.

Abstract: The paper describes a new approach to the optimal-power-flow problem based on Newton's method which it operates with an augmented Lagrangian function associated with the original problem. The function aggregates all the equality and inequality constraints. The first-order necessary conditions for optimality are reached by Newton's method, and by updating the dual variables and the penalty terms associated with the inequality constraints. The proposed approach does not have to identify the set of binding constraints and can be utilised for an infeasible starting point. The sparsity of the Hessian matrix of the augmented Lagrangian is completely exploited in the computational implementation. Tests results are presented to show the good performance of this approach.

Abstract: We describe a general approach decomposing a function into a sum of functions, each with a smaller input site than the original. Hence we can map such functions with essentially the same precision using small ROM tables and adders. We derive an easy method to compute the worst case error for many elementary functions and an error bound for the rest. Important applications are reciprocals, logarithms, exponentials and others. >

Abstract: Given a function f mappping n-variate inputs from a finite field F into F, we consider the task of reconstructing a list of all n-variate degree d polynomials which agree with f on a tiny but non-negligible fraction, /spl delta/, of the input space. We give a randomized algorithm for solving this task which accesses f as a black box and runs in time polynomial in 1//spl delta/, n and exponential in d, provided /spl delta/ is /spl Omega/(/spl radic/(d.

Abstract: The quasiclassical asymptotics of the Knizhnik-Zamolodchikov equation with values in the tensor product of sl(2)- representations are considered. The first term of asymptotics is an eigenvector of a system of commuting operators. We show that the norm of this vector with respect to the Shapovalov form is equal to the determinant of the matrix of second derivatives of a suitable function. This formula is an analog of the Gaudin and Korepin formulae for the norm of the Bethe vectors. We show that the eigenvectors form a basis under certain conditions.

Abstract: A computer system (10) and a method for determining a travel scheme minimizing travel costs for an organization, where the organization expects to purchase travel trips for a plurality of travelers for a plurality of travel links served by at least one carrier. The system comprises a data input device (54) for receiving travel information relating to the carriers and the links, a data storage device (34) for storing the received travel information, a processor (32) and a data output device (38). From the travel information, an objective function representing a travel cost for the travel trips and a set of constraints comprising restrictions relating to the function are constructed. A solution to the function is determined that satisfies the constraints and minimizes the travel cost. A report is generated by the data output device.

Abstract: The problem of minimizing a functional over a convex set of non-negative functions is considered, when the functional to be minimized is an f-entropy, or f-divergence resp. Bregman distance from a given function.

Abstract: Suppose that Q is a set of problems and S is a set of skills. A skill function assigns to each problem q i.e. to each element of Q — those sets of skills which are minimally sufficient to solve q; a problem function assigns to each set X of skills the set of problems which can be solved with these skills (a knowledge state). We explore the natural properties of such functions and show that these concepts are basically the same. Furthermore, we show that for every family K of subsets of Q which includes the empty set and Q, there are a set S of (abstract) skills and a problem function whose range is just K. We also give a bound for the number of skills needed to generate a specific set of knowledge states, and discuss various ways to supply a set of knowledge states with an underlying skill theory. Finally, a procedure is described to determine a skill function using coverings in partial orders which is applied to set A of the Coloured Progressive Matrices test (Raven, 1965).

Abstract: An effective logic synthesis procedure based on parallel and serial decomposition of a Boolean function is presented in this paper. The decomposition, carried out as the very first step of the .synthesis process, is based on an original representation of the function by a set of r-partitions over the set of minterms. Two different decomposition strategies, namely serial and parallel, are exploited by striking a balance between the two ideas. The presented procedure can be applied to completely or incompletely specified, single- or multiple-output functions and is suitable for different types of FPGAs including XILINX, ACTEL and ALGOTRONIX devices. The results of the benchmark experiments presented in the paper show that, in several cases, our method produces circuits of significantly reduced complexity compared to the solutions reported in the literature.

Abstract: We given anO(n logn)-time method for finding a bestk-link piecewise-linear function approximating ann-point planar point set using the well-known uniform metric to measure the error, ??0, of the approximation. Our methods is based upon new characterizations of such functions, which we exploit to design an efficient algorithm using a plane sweep in "? space" followed by several applications of the parametric-searching technique. The previous best running time for this problems wasO(n2).