论文信息 - Nonlinear classification by backprojection

Nonlinear classification by backprojection

A new method to construct a classification network, called the backprojection network, by learning from a given set of training exemplars is proposed. The method is derived from an analogy with the idea of image reconstruction by backprojection in computer-aided tomography. The backprojection network is able to correctly classify any distribution of training exemplars; can be incrementally constructed; has simple weights and low connectivity; and gives predictable generalization.<<ETX>>

Ping Liang

[1] J. Nadal,et al. Learning in feedforward layered networks: the tiling algorithm , 1989 .

[2] P. Liang. Problem decomposition and subgoaling in artificial neural networks , 1990, 1990 IEEE International Conference on Systems, Man, and Cybernetics Conference Proceedings.

[3] Ronald L. Rivest,et al. Training a 3-node neural network is NP-complete , 1988, COLT '88.

[4] K. Lang,et al. Learning to tell two spirals apart , 1988 .

[5] David Haussler,et al. What Size Net Gives Valid Generalization? , 1989, Neural Computation.

[6] Marcus Frean,et al. The Upstart Algorithm: A Method for Constructing and Training Feedforward Neural Networks , 1990, Neural Computation.

[7] M. Golea,et al. A Convergence Theorem for Sequential Learning in Two-Layer Perceptrons , 1990 .

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

[9] Hong Feng Yin,et al. A Connectionist Incremental Expert System Combining Production Systems and Associative Memory , 1991, Int. J. Pattern Recognit. Artif. Intell..