Title :
Layered inductive learning algorithms and their computational aspects
Abstract :
Inductive learning algorithms are based on artificial neural network structures and on the principle of self-organization. The inductive approach is realized by enumerating a large number of equations to choose the best. Linear combinations of inputs are generated in the network and are evaluated according to the changes in response to a comparison with externally fixed criteria, which is adapted as a learning mechanism. This means that outputs of some units are discarded at various hierarchical levels of the network structure. Algorithms (multilayer, combinatorial, and harmonic) based on this mechanism differ in their network structures in generating partial functions and in using various heuristics. A system modeling technique is to be used for estimating the coefficients of the functions at each neuron cell or unit. The least-squares method is demonstrated in these networks
Keywords :
learning systems; neural nets; self-adjusting systems; artificial neural network structures; coefficients estimation; combinatorial algorithms; computational aspects; equations; externally fixed criteria; harmonic algorithms; heuristics; hierarchical levels; layered inductive learning algorithms; least-squares method; multilayer algorithms; partial functions; self-organization; system modeling technique; Biological neural networks; Brain modeling; Computer networks; Fires; Integrated circuit interconnections; Learning systems; Multi-layer neural network; Nerve fibers; Neural networks; Neurons;
Conference_Titel :
Tools for Artificial Intelligence, 1989. Architectures, Languages and Algorithms, IEEE International Workshop on
Conference_Location :
Fairfax, VA
Print_ISBN :
0-8186-1984-8
DOI :
10.1109/TAI.1989.65353
