The availability of efficient software implementations of neural network algorithms is a key task in the development phase to evaluate the network results in real cases. In this letter we address the problem of optimizing the sequential algorithm for the Boltzmann Machine (BM). We present a solution which is based on the locality properties of the algorithm and enables a very efficient computation of the cost difference between two configurations. Since the algorithm performance depends on the number of accepted state transitions in the annealing process, we formulate a theoretical procedure to estimate the acceptance probability of a state transition. In addition, we provide experimental data on a well-known optimization problem (TSP) to have a numerical verification of the theory, and to show that the proposed solution obtains a speedup between 3 and 4 in comparison with the traditional algorithm.
EFFICIENT IMPLEMENTATION OF THE BOLTZMANN MACHINE ALGORITHM / A., De Gloria; P., Faraboschi; Olivieri, Mauro. - In: IEEE TRANSACTIONS ON NEURAL NETWORKS. - ISSN 1045-9227. - 4:1(1993), pp. 159-163. [10.1109/72.182711]
EFFICIENT IMPLEMENTATION OF THE BOLTZMANN MACHINE ALGORITHM
OLIVIERI, Mauro
1993
Abstract
The availability of efficient software implementations of neural network algorithms is a key task in the development phase to evaluate the network results in real cases. In this letter we address the problem of optimizing the sequential algorithm for the Boltzmann Machine (BM). We present a solution which is based on the locality properties of the algorithm and enables a very efficient computation of the cost difference between two configurations. Since the algorithm performance depends on the number of accepted state transitions in the annealing process, we formulate a theoretical procedure to estimate the acceptance probability of a state transition. In addition, we provide experimental data on a well-known optimization problem (TSP) to have a numerical verification of the theory, and to show that the proposed solution obtains a speedup between 3 and 4 in comparison with the traditional algorithm.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
