Simulated annealing and Boltzmann machines: a stochastic approach to combinatorial optimization and neural computing
IEEE Transactions on Computers - Special issue on artificial neural networks
An efficient mapping of Boltzmann Machine computations onto distributed-memory multiprocessors
Microprocessing and Microprogramming
Relaxing Synchronization in Distributed Simulated Annealing
IEEE Transactions on Parallel and Distributed Systems
Synchronous and Asynchronous Parallel Simulated Annealing with Multiple Markov Chains
IEEE Transactions on Parallel and Distributed Systems
An Inherently Parallel Method for Heuristic Problem-Solving: Part I-General Framework
IEEE Transactions on Parallel and Distributed Systems
An asynchronous distributed architecture model for the Boltzmann machine control mechanism
IEEE Transactions on Neural Networks
Analyzing Boltzmann Machine Parameters for Fast Convergence
IWANN '01 Proceedings of the 6th International Work-Conference on Artificial and Natural Neural Networks: Bio-inspired Applications of Connectionism-Part II
Hi-index | 0.00 |
The resolution of combinatorial optimization problemscan greatly benefit from the parallel and distributedprocessing which is characteristic of neural networkparadigms. Nevertheless, the fine grain parallelism ofthe usual neural models cannot be implemented in anentirely efficient way either in general-purposemulticomputers or in networks of computers, which arenowadays the most common parallel computerarchitectures. Therefore, we present a parallelimplementation of a modified Boltzmann machine wherethe neurons are distributed among the processors ofthe multicomputer, which asynchronously compute theevolution of their subset of neurons using values forthe other neurons that might not be updated, thusreducing the communication requirements. Severalalternatives to allow the processors to workcooperatively are analyzed and their performancedetailed. Among the proposed schemes, we haveidentified one that allows the corresponding BoltzmannMachine to converge to solutions with high quality andwhich provides a high acceleration over the executionof the Boltzmann machine in uniprocessor computers.