Kishan G. Mehrotra

Learn More
The rate of convergence of net output error is very low when training feedforward neural networks for multiclass problems using the backpropagation algorithm. While backpropagation will reduce the Euclidean distance between the actual and desired output vectors, the differences between some of the components of these vectors increase in the first iteration.(More)
Abstract--This pal~er presents a neural network approach to multivariate time-series anal.lwis. Real world observations ~?/flottr /.'ices in three cities have been used as a benchmark in our ev~eriments, l"eed/orward connectionist networks have bt,t,tl ~k,.signed to model flottr l , ices over the period/iom ..lltgttsl 1972 to Novt,mher 1980./or the cities(More)
The backpropagation algorithm converges very slowly for two-class problems in which most of the exemplars belong to one dominant class. An analysis shows that this occurs because the computed net error gradient vector is dominated by the bigger class so much that the net error for the exemplars in the smaller class increases significantly in the initial(More)
We present practical parallel algorithms using prefix computations for various problems that arise in pairwise comparison of biological sequences. We consider both constant and affine gap penalty functions, full-sequence and subsequence matching, and space-saving algorithms. The best known sequential algorithms solve these problems in O(mn) time and O(m +(More)
Thispaperpresents aperformance evaluation approach to compare different distributed load balancing schemes on a unified basis. This approach is an integration of simulation, statistical and analytical models, and takes into account thefundamental system parameters that can possibly affect the pe~ormance. We show that all the sender+”nitiated distributed(More)
This paper presents three approaches to improve fault tolerance of neural networks. In two approaches , the traditional backpropagation training algorithm is itself modiied so that the trained networks have improved fault tolerance; we achieve better results than others 1, 10] who had also explored this possibility. Our rst method is to coerce weights to(More)
Partitioning graphs into equally large groups of nodes, minimizing the number of edges between different groups, is an extremely important problem in parallel computing. This paper presents genetic algorithms for suboptimal graph partitioning, with new crossover operators (KNUX, DKNUX) that lead to orders of magnitude improvement over traditional genetic(More)
An efficient maximum-likelihood soft-decision decoding algorithm for linear block codes using a generalized Dijkstra’s algorithm was proposed by Han, Hartmann, and Chen. In this correspondence we prove that this algorithm is efficient for most practical communication systems where the probability of error is less than 10 3 by finding an upper bound of the(More)
The goal of this dissertation is to develop a paradigm for the next generation of software applications with a clear architecture that unifies desktop and Internet applications. It is aimed at addressing the issues of leveraging existing software assets and incorporating advanced capabilities including collaboration and universal access. As the overall Web(More)