Coherence in the Visual Imagination: Local Hill Search Outperforms Thagard’s Connectionist Model

Michael VertolliCarleton University, Ottawa, Ontario, Canada
Jim DaviesCarleton University, Ottawa, Ontario, Canada

Abstract

A cognitive model of the visual imagination will produce “incoherent” results when it adds elements to an imagined scene that come from different contexts (e.g., “computer” and “cheese” with “mouse”). We approach this problem with a model that infers coherence relations from co-occurrence probabilities of labels in images. We show that this algorithm’s serial traversal of networks of co-occurrence relations for a particular query produces greater coherence than one leading model in the field of computational coherence: Thagard’s connectionist model.

Files

Coherence in the Visual Imagination: Local Hill Search Outperforms Thagard’s Connectionist Model (749 KB)



Back to Table of Contents