论文信息 - Implementing recurrent back-propagation on the connection machine

Implementing recurrent back-propagation on the connection machine

Abstract The recurrent back-propagation algorithm for neural networks has been implemented on the Connection Machine, a massively parallel processor. Two fundamentally different graph architectures underlying the nets were tested: one based on arcs, the other on nodes. Confirming the predominance of communication over computation, performance measurements underscore the necessity to make connections the basic unit of representation. Comparisons between these graph algorithms lead to important conclusions concerning the parallel implementation of neural nets in both software and hardware.

Etienne Deprit

[1] Alan S. Lapedes,et al. A self-optimizing, nonsymmetrical neural net for content addressable memory and pattern recognition , 1986 .

[2] Fernando J. Pineda,et al. Dynamics and architecture for neural computation , 1988, J. Complex..

[3] S. Tomboulian. Overview and extensions of a system for routing directed graphs on SIMD architectures , 1988, Proceedings., 2nd Symposium on the Frontiers of Massively Parallel Computation.

[4] Pineda,et al. Generalization of back-propagation to recurrent neural networks. , 1987, Physical review letters.

[5] Luís B. Almeida,et al. A learning rule for asynchronous perceptrons with feedback in a combinatorial environment , 1990 .

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

[7] Sherryl Jean Tomboulian. A system for routing arbitrary communication graphs on simd architectures , 1986 .

[8] W. Daniel Hillis,et al. The connection machine , 1985 .

[9] A. E. Vries,et al. Separation of 14C16O and 12C18O by thermal diffusion , 1956 .

[10] W. Daniel Hillis,et al. Data parallel algorithms , 1986, CACM.

[11] Guy E. Blelloch,et al. An implementation of network learning on the Connection Machine , 1988 .