UNB Scholar Research Repository :: Browsing by Author "Bhavsar, Virendra, C."

Browsing by Author "Bhavsar, Virendra, C."

Now showing 1 - 15 of 15

A comparison of vectorizable discrete sampling methods in Monte Carlo applications
(1995) Sarno, Riyanarto; Bhavsar, Virendra, C.; Hussein, Esam, M., A.
The performance of various vectorizable discrete random-sampling methods, along with the commonly used inverse sampling method, is assessed on a vector machine. Monte Carlo applications involving, one-dimensional, two-dimensional and multi-dimensional probability tables are used in the investigation. Various forms of the weighted sampling method and methods that transform the original probability table are examined. It is found that some form of weighted sampling is efficient, when the original probability distribution is not far from uniform or can be approximated analytically. Table transformation methods, though require additional memory storage, are best suited in applications where multi-dimensional tables are involved. Keywords: Discrete sampling, Weighted sampling, Monte Carlo simulations, Vector processing.
A New Approach for Specification and Verification Of Distributed Agents
(2000) Mironov, Andrew, M.; Bhavsar, Virendra, C.
Adaptive resonance theory networks using incremental communication
Chen, Ming; Ghorbani, Ali, A.; Bhavsar, Virendra, C.
The incremental inter-node communication method is applied to the adaptive resonance theory 2 (ART2) networks. The incremental communication is aimed at reducing the communication costs of parallel and VLSI implementations of artificial neural networks. A node architecture incorporating the incremental communication is presented. A simulator is developed to stndy the behavior of ART2 networks with varying precisions of incremental data communication. Experiments are carried out to study the effects of the incremental communication on the convergence and savings in communication costs. We have found that even 7-bit precision in fixed-point and 13-bit (including 8-bit exponent) floating-point representations may be sufficient for the network to give the same results as those with conventional communication using 32-bit precision. The simulation results show that the limited precision errors are bounded and do not seriously affect the convergence of ART2 networks.
Analysis and synthesis of interpolating functions for 3-D objects
(1986) Datar, Narendra, N.; Gujar, Uday, G.; Bhavsar, Virendra, C.
Interpolation technique has been used for generation of three dimensional surfaces. This paper first analyses the objects generated by specified interpolating functions. Both linear and non-linear interpolating functions are used to generate interesting objects. None of these objects can be generated using the technique of surface of revolution. The paper discusses various choices for the interpolating weighting functions and introduces the idea of modifying functions to incorporate second order of non-linearity. Several interesting objects, such as those resembling chess pieces, vase, goblets, etc., are presented. The main feature of the generated objects is that they have a compact mathematical representation. Subsequently, we consider the reverse process where the characteristics of the desired shape are given and the interpolating functions are to be determined. The Fourier series expansion is used to determine an interpolating function which gives rise to the desired object using non-linear interpolation. This synthesis process is illustrated with several examples. CR Categories and Subject Descriptors: G.1.1 [Numerical Analysis]: Interpolation - Linear and Nonlinear, Fourier Series Analysis, I.3.5 [Computer Graphics]: Computational Geometry and Object Modeling - curve, surface, sold and object representation. General Terms: Analysis and Synthesis, Generation of 3-D Objects. Additional Keywords and Phrase: inbetweening, blending and modifying functions.
COMMAN: a communication analyzer for OCCAM 2
(1994) d'Auriol, Brian, J.; Bhavsar, Virendra, C.
We present a parser for analyzing communication overheads in OCCAM 2 programs for transputers. The input to COMMAN is a source file containing PROTOCOL statements which are defined by an augmented OCCAM 2 grammar. The output of the parser is a report detailing the expected communication times for the given input protocols. We show that COMMAN reasonably predicts transputer communication times for a variety of modes including internal and external communication, using both sequential as well as variant protocols. The parser is implemented using yacc while the lexical scanner is implemented using lex, which are utilities available in the unix and other environments. We give details of the lex and yacc implementations including the tokens recognized, the detection of scope and the grammar production rules. The communication times are modeled by linear functions and are built into the yacc grammar production rules. Keywords: Communication Analysis, OCCAM 2, parsing, Transputer, multicomputers, parallel processing
Design and analysis of vectorized Monte Carlo codes
(1989) Sarno, Riyanarto; Bhavsar, Virendra, C.; Banerjee, Pradeep, K.
Vectorized Monte Carlo codes use a set of vectors, generally referred to as a stack in the literature, to hold the attribute values of the entities carrying out random walks. This paper presents schemes for optimizing the performance of stack processing in Monte Carlo codes and carries out their execution time analyses. The proposed four schemes are: (i) continuous inspection of the stack with no stack compression, (ii) continuous inspection of the stack with stack compression, (iii) periodical inspection of the stack with no stack compression, and (iv) periodical inspection of the stack with stack compression. The execution time analysis of the continuous schemes is carried out using some results from Order Statistics, and that of the periodical inspection schemes is carried out using some results from Markovian decision processes. Under some assumptions, one of them being that the time required for random walk computations is exponentially distributed, closed-form expressions for the expected execution time are obtained. The theoretical performance of the proposed schemes is illustrated and concluding remarks are given. Finally, the performance of the schemes is illustrated through vectorizing few Monte Carlo codes and running them on IBM 3090-180VF. Keywords: Markovian decision processes, Monte Carlo codes, order statistics, parallel processing, supercomputing, vector processing
Evaluation of the discrepancy of the linear congruential pseudo-random number sequences
(1987) Bhavsar, Virendra, C.; Lambrou, Lambros, A; Horton, Joseph, D; Gujar, Uday, G
Niederreiter has defined the discrepancy of a pseudo-random number (PRN) sequence as a quantity which measures the deviation of the sequence's distribution from the ideal uniform distribution. The discrepancy evaluation appears to be a promising criteria for testing randomness of PRN sequences. In this paper, we propose two algorithms for computer evaluation of the discrepancy of PRN sequences and evaluate the discrepancy of PRN sequences generated by a linear congruential method.
Fractal images from z <- za + c in the complex c-plane
(1988) Gujar, Uday, G.; Bhavsar, Virendra, C.
In this paper, we propose the generalized transformation function z <- z[superscript a] + c for generating fractal images. The self-squared function z <- z[superscript 2] + c, which is discussed extensively in the literature, is a special case of this function. A multitude of interesting, intriguing and rich families of fractals are generated by changing a single parameter, a. Direct relationships are observed between a and the visual characteristics of the fractal image in the c-plane. The exponent a can be represented as a = +-(n+e), where n and e are the integer and fractional parts, respectively. It is found that when a is a positive integer number, the resulting image contains lobular structures. The number of major lobes equals (n-1). When a is a negative integer number, the generated fractal image is a planetary structure consisting of overlapping central planets surrounded by satellite structures. The number of satellite structures equals (n+1). A continuous variation of a between two consecutive integers results into a continuous proportional change between the two limiting fractal images. Several conjectures about the visual characteristics of the images and the value of a are stated.
Geometric fractals using traversal strategies
(1992) Gujar, Uday, G.; Bhavsar, Virendra, C.; Choi, Stephen, Y. M.; Kalra, Prem, K.
In the literature it is assumed that given a line segment and a generator there is only one way to replace the line segment by the generator to produce the fractal image, i.e., a fixed traversal direction is assumed for each segment. We introduce the concept of traversal strategies wherein each of the segments of the initiator and generator can be traversed in three ways. This results in 3M • 3Q fractal images for a Q-segment initiator and an M-segment generator; these are referred to as the traversed geometric fractals. In contrast, a single fractal image is obtained without the use of traversal strategies. A software tool, Rangoli, has been developed to generate traversed geometric fractals. Several examples are included. Keywords: Geometric fractals, traversal strategies, generators, initiators, fractal images.
Incremental communication for artificial neural networks
(1993) Ghorbani, Ali, A.; Bhavsar, Virendra, C.
A learning procedure based on the backpropagation algorithm using the incremental communication is presented. In the incremental communication method instead of communicating the whole value of a variable, the increment or decrement to its previous value is only sent on a communication link. The incremental value may be either a fixed-point or a floating-point value. The method is applied to four different error backpropagation networks and the effect of the precision of the incremental values of activation, weights and error signals on the convergence behavior is examined. It is shown through simulation that at least 7-bit precision in fixed-point and 2-digit precision in floating-point representations are required for the network to generalize. With 12-bit fixed-point or 4-digit floating-point precision almost the same results are obtained as that with the conventional communication using 32-bit precision. The proposed method of communication can lead to enormous savings in the communication cost for implementations of artificial neural networks on parallel computers as well as direct hardware realizations. This method is applicable to many other types of artificial neural systems and can be incorporated along with the other limited precision strategies for representation of variables suggested in literature.
Inductive theory of vision
(1996) Goldfarb, Lev; Deshpande, Sanjay, S.; Bhavsar, Virendra, C.
In spite of the fact that some of the outstanding physiologists and neurophysiologists (e.g. Hermann von Helmholtz and Horace Barlow) insisted on the central role of inductive learning processes in vision as well as in other sensory processes, there are absolutely no (computational) theories of vision that are guided by these processes. It appears that this is mainly due to the lack of understanding of what inductive learning processes are. We strongly believe in the central role of inductive learning processes, around which, we think, all other (intelligent) biological processes have evolved. In this paper we outline a (computational) theory of vision completely built around the inductive learning processes for all levels in vision. The development of such a theory became possible with the advent of the formal model of inductive learning--evolving transformation system (ETS). The proposed theory is based on the concept of structured measurement device, which is motivated by the formal model of inductive learning and is a far-reaching generalization of the concept of classical measurement device whose output measurements are not numbers but structured entities ("symbols") with an appropriate metric geometry. We propose that the triad of object structure, image structure and the appropriate mathematical structure (ETS)--to capture the latter two structures-is precisely what computational vision should be about. And it is the inductive learning process that relates the members of this triad. We suggest that since the structure of objects in the universe has evolved in a combinative (agglomerative) and hierarchical manner, it is quite natural to expect that biological processes have also evolved (to learn) to capture the latter combinative and hierarchical structure. In connection with this, the inadequacy of the classical mathematical structures as well as the role of mathematical structures in information processing are discussed. We propose the following postulates on which we base the theory. Postulate 1. The objects in the universe have emergent combinative hierarchical structure. Moreover, the term "object structure" cannot be properly understood and defined outside the inductive learning process. Postulate 2. The inductive learning process is an evolving process that tries to capture the emergent object (class) structure mentioned in Postulate 1. The mathematical structure on which the inductive learning model is based should have the intrinsic capability to capture the evolving object structure. (It turns out that the corresponding mathematical structure is fundamentally different from the classical mathematical structures.) Postulate 3. All basic representations in vision processes are constructed on the basis of the inductive image representation, which, in turn, is constructed by the inductive learning process (see Postulate 2). Thus, the inductive learning processes form the core around which all vision processes have evolved. We present simple examples to illustrate the proposed theory for the case of "low-level" vision.
Interpolation techniques for 3-D object generation
(1987) Gujar, Uday, G.; Bhavsar, Virendra, C.; Datar, Narendra, N.
The technique of interpolation (or blending) and its generalization is considered in this paper. A unifying model for generating 3-D objects using this technique is presented. Various factors which affect the shape of generated objects are identified. Several illustrative examples of 3-D objects generated by this technique, using linear as well as non-linear interpolation, are presented. Although many of the objects have complex shapes, they are represented by simple closed form mathematical equations.
Multi-transputer implementations of selected image processing techniques
(1993) Turner, Christopher, J.; Bhavsar, Virendra, C.; Pochec, Przemyslaw, R.
We present parallel implementations of a local (convolution) and a global (regular moment extraction) image processing technique on a multi-transputer system. Issues relevant to implementation design including computational algorithm selection, initial data pass, and topology selection are discussed. Linear speedups in the convolution implementations are observed for all image sizes whereas such speedups are obtained only for large image sizes for the regular moment implementations. Analysis of the implementations including parallel time complexity functions and observations about data passing and topology selection is given. Two theoretical performance models based on the implementations closely match empirical timing results. Keywords: parallel processing, transputers, image processing, convolution, moments, performance evaluation.
Parallel tangent learning algorithm for training artificial neural networks
(1993) Ghorbani, Ali, A.; Bhavsar, Virendra, C.
A modified backpropagation training algorithm using deflecting gradient technique is proposed. Parallel tangent (Partan) gradient is used as a deflecting method to accelerate the convergence. This method can also be thought as a particular implementation of the method of conjugate gradient. Partan gradient consists of two phases namely, climbing through gradient and accelerating through parallel tangent. Partan overcomes the inefficiency of zigzagging in the conventional backpropagation learning algorithm by deflecting the gradient through acceleration phase. The effectiveness of the proposed method in decreasing the rate of convergence is investigated by applying it to four learning problems with different error surfaces. It is found through simulation that regardless of the degree of the complexity of the problems used, the Partan backpropagation algorithm shows faster rate of convergence to the solution. In particular, for the exclusive-or problem its convergence time is approximately five times faster than that of standard backpropagation, whereas about two times faster rate of convergence is obtained for Encoder/Decoder, Binary-to-local, and Sonar problems.
The ACORN Multi-Agent System
(2001) Marsh, Stephen; Ghorbani, Ali; Bhavsar, Virendra, C.
ACORN (Agent-based Community Oriented Routing Network) is a distributed multi-agent architecture for the search, distribution and management of information across networks. ACORN utilises the concept of 'information as agent' together with an application of Stanley Milgram's Small World Problem (the idea of Six Degrees of Separation) in order to route individual items of information around a network of people and agents. The ACORN ideal is to achieve a state where a web of users is created such that information distribution, queries and search, and browsing behaviour is encapsulated in a single adaptive architecture which learns community behaviour and knowledge in order to route agents to relevant destinations (users). This paper describes the ACORN architecture and its implementation. We introduce a novel idea of agent meeting places, or Cafes, to carry out community-based information sharing among mobile agents in ACORN. ACORN is compared with similar work, and evaluations of ACORN for information sharing among mobile agents are described. Applications of ACORN include Business to Business and Business to Consumer based e-Commerce solutions, virtual community creation and support systems, peer reviewing systems, and personalized directed information handling. Keywords: Multi-Agent information architectures, autonomous agents, mobile agents, keyphrase matching, multi-agent architecture, community based information handling, e-Commerce.

Browsing by Author "Bhavsar, Virendra, C."

Results Per Page

Sort Options