Inductive machine learning bias in knowledge-based neurocomputing

[摘要] ENGLISH ABSTRACT:The integration of symbolic knowledge with artificial neural networks is becoming anincreasingly popular paradigm for solving real-world problems. This paradigm namedknowledge-based neurocomputing, provides means for using prior knowledge to determinethe network architecture, to program a subset of weights to induce a learning biaswhich guides network training, and to extract refined knowledge from trained neuralnetworks. The role of neural networks then becomes that of knowledge refinement. Itthus provides a methodology for dealing with uncertainty in the initial domain theory.In this thesis, we address several advantages of this paradigm and propose a solutionfor the open question of determining the strength of this learning, or inductive, bias.We develop a heuristic for determining the strength of the inductive bias that takes thenetwork architecture, the prior knowledge, the learning method, and the training datainto consideration.We apply this heuristic to well-known synthetic problems as well as published difficultreal-world problems in the domain of molecular biology and medical diagnoses. Wefound that, not only do the networks trained with this adaptive inductive bias showsuperior performance over networks trained with the standard method of determiningthe strength of the inductive bias, but that the extracted refined knowledge from thesetrained networks deliver more concise and accurate domain theories.

[发布机构] Stellenbosch University