Abstract: We develop an approach to analogical reasoning with hierarchically structured descriptions of episodes and situations based on a particular form of vector representations – structure-sensitive sparse binary distributed representations known as code-vectors. We propose distributed representations of analog elements that allow finding correspondence between the elements for implementing analogical mapping, as well as analogical inference, based on similarity of those representations. The proposed methods are investigated using test analogs and the obtained results are as those of known mature analogy models. However, exploiting similarity properties of distributed representations provides a better scaling, enhances the semantic basis of analogs and their elements as well as neurobiological plausibility. The paper also provides a brief survey of analogical reasoning, its models, and representations employed in those models.

Keywords: analogy, analogical mapping, analogical inference, distributed representation, code-vector, reasoning, knowledge bases.

ACM Classification Keywords: I.2 ARTIFICIAL INTELLIGENCE, I.2.4 Knowledge Representation Formalisms and Methods, I.2.6 Learning (Analogies)

Link:

ANALOGICAL MAPPING USING SIMILARITY OF BINARY DISTRIBUTED REPRESENTATIONS

Serge V. Slipchenko, Dmitri A. Rachkovskij

http://foibg.com/ijita/vol16/IJITA16-3-p07.pdf