摘要: A feedforward network is proposed which lends itself to cost-effective implementations in digital hardware and has a fast forward-pass capability. It differs from the conventional model restricting its synapses set {-1, 0, 1} while allowing unrestricted offsets. Simulation results on 'onset of diabetes' data handwritten numeral recognition database indicate that new network, despite having strong constraints synapses, generalization performance similar counterpart.
altafkhan.com 参考文章(8)
Guyon, Poujaud, Personnaz, Dreyfus, Denker, Le Cun, Comparing different neural network architectures for classifying handwritten digits international joint conference on neural network. pp. 127- 132 ,(1989) , 10.1109/IJCNN.1989.118570
Halbert White, Artificial Neural Networks: Approximation and Learning Theory ,(1992)
Halbert White, Learning in Artificial Neural Networks: A Statistical Perspective Neural Computation. ,vol. 1, pp. 425- 464 ,(1989) , 10.1162/NECO.1989.1.4.425
A.H. Khan, E.L. Hines, Integer-weight neural nets Electronics Letters. ,vol. 30, pp. 1237- 1238 ,(1994) , 10.1049/EL:19940817
PeterH Bennett, ThomasA Burch, Max Miller, DIABETES MELLITUS IN AMERICAN (PIMA) INDIANS The Lancet. ,vol. 298, pp. 125- 128 ,(1971) , 10.1016/S0140-6736(71)92303-8
H. Drucker, Y. Le Cun, Improving generalization performance using double backpropagation IEEE Transactions on Neural Networks. ,vol. 3, pp. 991- 997 ,(1992) , 10.1109/72.165600
N. L. Hjort, Brian D. Ripley, Pattern recognition and neural networks ,(1996)
C.-M. Kuan, K. Hornik, Convergence of learning algorithms with constant learning rates IEEE Transactions on Neural Networks. ,vol. 2, pp. 484- 489 ,(1991) , 10.1109/72.134285