Abstract
The overwhelming majority of research currently pursued within the framework of concept-learning concentrates on discrimination-based learning, an inductive learning paradigm that relies on both examples and counter-examples of the concept. This emphasis, however, can present a practical problem: there are real-world engineering problems for which counter-examples are both scarce and difficult to gather. For these problems, recognition-based learning systems are much more appropriate because they do not use counter-examples in the concept learning phase. The purpose of this dissertation is to analyze a connectionist recognition-based learning system|autoassociation-based classication|and answer the following questions: What features of the auto associator make it capable of performing classic- cation in the absence of counter-examples? What causes the auto associator to be significantly more efficient than MLP in certain domains? What domain characteristics cause the auto associator to be more accurate than MLP and MLP to be more accurate than the auto associator? The dissertation concludes that 1) auto association-based classification is possible in a particular class of practical domains called non-linear and multi-modal because the auto associator uses a multi-modal specialization bias to compensate for the absence of counter-examples. This bias can be controlled by varying the capacity of the auto associator. 2) The difference in efficiency between the auto associator and MLP observed on this class of domains is caused by the fact that the auto associator uses a (fast) data-driven generalization strategy whereas MLP has recourse to a (slow) hypothesis-driven one, despite the fact that the two systems are both trained by the backpropagation procedure. 3) The auto associator classifies more accurately than MLP domains requiring particularly strong specialization biases caused by the counter-conceptual class or particularly weak specialization biases caused by the conceptual class. However, MLP is more accurate than the auto associator on domains requiring particularly strong specialization biases caused by the conceptual class. The results of this study thus suggest that recognition-based systems, which are often dismissed in favor of discrimination-based ones in the context of concept-learning, may present an interesting array of classification strengths.