A Proposal for More Powerful Learning Algorithms

被引:27
作者
Baum, Eric B. [1 ]
机构
[1] CALTECH, Jet Prop Lab, Pasadena, CA 91109 USA
基金
美国国家航空航天局;
关键词
D O I
10.1162/neco.1989.1.2.201
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Judd (1988) and Blum and Rivest (1988) have recently proved that the loading problem for neural networks is NP complete. This makes it very unlikely that any algorithm like backpropagation which varies weights on a network of fixed size and topology will be able to learn in polynomial time. However, Valiant has recently proposed a learning protocol (Valiant 1984) which allows one to sensibly consider generalization by learning algorithms with the freedom to add neurons and synapses, as well as simply adjusting weights. Within this context, standard circuit complexity arguments show that learning algorithms with such freedom can solve in polynomial time any learning problem that can be solved in polynomial time by any algorithm whatever. In this sense, neural nets are universal learners, capable of learning any learnable class of concepts.
引用
收藏
页码:201 / 207
页数:7
相关论文
共 15 条
[1]  
Baum E. B., 1988, Journal of Complexity, V4, P193, DOI 10.1016/0885-064X(88)90020-9
[2]  
Baum E. B., 1988, COMPLEXITY IN PRESS
[3]   What Size Net Gives Valid Generalization? [J].
Baum, Eric B. ;
Haussler, David .
NEURAL COMPUTATION, 1989, 1 (01) :151-160
[4]  
BLUM A, 1988, P 1 ANN WORKSH COMP, P00009
[5]  
Blumer A., 1987, J ASS COMPU IN PRESS
[6]  
Goldreich O., 1984, J ASSOC COMPUT MACH, V33, P4
[7]  
Judd S., 1988, J COMPLEXITY, P4
[8]   A NEW POLYNOMIAL-TIME ALGORITHM FOR LINEAR-PROGRAMMING [J].
KARMARKAR, N .
COMBINATORICA, 1984, 4 (04) :373-395
[9]  
Kearns M., 1988, 1488 HARV U AIK LAB
[10]   RELATIONS AMONG COMPLEXITY MEASURES [J].
PIPPENGER, N ;
FISCHER, MJ .
JOURNAL OF THE ACM, 1979, 26 (02) :361-381