The main contribution of this work is the proposal of a system that attempts to combine inductive
learning with prior knowledge and reasoning in a unified framework. The model described in [22, 23]
is a connectionist approach to implementing the form of reasoning discussed above (with forward
chaining). However, this model only deals with representational issues, and not learning. ILA's
reasoning power is more limited, but the system is able to learn from examples. ILA's inductive
learning is similar to NGE [20], but ILA uses hyperplanes rather than hyperrectangles as generalized
exemplars. ILA also bears some resemblance with IBL [2], in the similarity measure used. However,
ILA also incorporates prior knowledge and must provide mechanisms to support reasoning. Finally, if
no rules are induced from the examples in learning, then ILA degenerates to a simplified form of MBR
[21].
Section 2 introduces ILA, presents an example, and gives a comparison of ILA with other related
work. Section 3 presents several simulation results on many real-world datasets. Section 4 discusses
each of the above issues in detail, as well as some of the difficulties associated with ILA, such as the
ordering of information in learning. Finally, section 5 concludes the paper.