Implementation of ANN on RISC processor array

The authors present a mesh systolic array, GCN (giga connection), for a fast simulator of artificial neural networks (ANNs). The processor element (PE) of the GCN is composed of the RISC processor i-860 designed by Intel Corp., a large scale local memory, and high bandwidth first-in first-out devices. The mapping algorithm of the ANN onto the GCN, called the net-data partition, is discussed, and the multilayer feedforward network and Kohenen feature map are mapped onto the GCN by using this algorithm. Another parallelism that can be used for a stochastic ANN like the Boltzmann machine is also discussed. The performance of the GCN is evaluated by software simulation and the authors achieve over 1 gigaconnection per second using 128 PEs.<<ETX>>

[1]  Teuvo Kohonen,et al.  Self-Organization and Associative Memory , 1988 .

[2]  B. Bavarian,et al.  An ART2-TPM neural network for automatic pattern classification , 1991, IJCNN-91-Seattle International Joint Conference on Neural Networks.

[3]  C. D. Gelatt,et al.  Optimization by Simulated Annealing , 1983, Science.

[4]  D. S. Touretzky,et al.  Neural network simulation at Warp speed: how we got 17 million connections per second , 1988, IEEE 1988 International Conference on Neural Networks.

[5]  Geoffrey E. Hinton,et al.  Learning internal representations by error propagation , 1986 .

[6]  H. T. Kung,et al.  The Warp Computer: Architecture, Implementation, and Performance , 1987, IEEE Transactions on Computers.

[7]  Alex Waibel,et al.  Phoneme recognition: neural networks vs. hidden Markov models vs. hidden Markov models , 1988, ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing.

[8]  Donald Geman,et al.  Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[9]  Geoffrey E. Hinton,et al.  Learning and relearning in Boltzmann machines , 1986 .