Addressing knowledge-representation issues in connectionist symbolic rule encoding for general inference

摘要

This chapter describes one method for addressing knowledge representation issues that arise when a connectionist system replicates a standard symbolic style of inference for general inference. Symbolic rules are encoded into the networks, called structured predicate networks (SPN) using neuron-like elements. Knowledge-representation issues such as unification and consistency checking between two groups of unifying arguments arise when a chain of inference is formed over the networks encoding special type of symbol rules. These issues are addressed by connectionist sub-mechanisms embedded into the networks. As a result, the proposed SPN architecture is able to translate a significant subset of first-order Horn Clause expressions into a connectionist representation that may be executed very efficiently.