Constructive Induction Approach to Growing Neural Networks - GMU Machine Learning and Inference Laboratory

constructive induction approach to growing neural networks

(Sazonow, Wnek)

With most symbolic machine learning methods, if the given knowledge representation space is inadequate then the learning process will fail. This is also true with neural networks learning based methods. To overcome this limitation, a method for automatically "growing" neural network is being developed.

The BP-HCI method is a hypothesis-driven constructive induction for neural networks trained by the back propagation algorithm. The method determines topology of a neural network and the initial connection weights based on patterns in the behavior of the neural network. The behavior of the neural network is captured by concepts called ACCORD and ANXIETY of a neural network.

The method was successfully applied to ten problems including such problems as learning "exclusive-or" function, MONK2, parity-6BIT and inverse parity-6BIT.

Selected References

Sazonov, V.N. and Wnek, J., "Hypothesis-driven Constructive Induction Approach to Expanding Neural Networks," Working Notes of the ML-COLT'94 Workshop on Constructive Induction and Change of Representation, New Brunswick, NJ, July 1994.

For more references, seeĀ publications section.

MLI Copyright © 2017 Machine Learning and Inference Laboratory
College of Health and Human Services, George Mason University
4400 University Dr, MSN 1J3, Fairfax, VA 22030, U.S.A