Modeling visual problem solving as analogical reasoning.

Abstract

We present a computational model of visual problem solving, designed to solve problems from the Raven's Progressive Matrices intelligence test. The model builds on the claim that analogical reasoning lies at the heart of visual problem solving, and intelligence more broadly. Images are compared via structure mapping, aligning the common relational structure in 2 images to identify commonalities and differences. These commonalities or differences can themselves be reified and used as the input for future comparisons. When images fail to align, the model dynamically rerepresents them to facilitate the comparison. In our analysis, we find that the model matches adult human performance on the Standard Progressive Matrices test, and that problems which are difficult for the model are also difficult for people. Furthermore, we show that model operations involving abstraction and rerepresentation are particularly difficult for people, suggesting that these operations may be critical for performing visual problem solving, and reasoning more generally, at the highest level. (PsycINFO Database Record

RAVEN: A Dataset for Relational and Analogical Visual REasoNing

“…Computational Efforts in RPM The research community of cognitive science has tried to attack the problem of RPM with computational models earlier than the computer science community. However, an oversimplified assumption was usually made in the experiments that the computer programs had access to a symbolic representation of the image and the operations of rules [7,32,33,34]. As reported in Section 4.4, we show that giving this critical information essentially turns it into a searching prob-lem.…”

“…These representations derived from the A-SIG allow a new form of reasoning, i.e., one that combines visual understanding and structure reasoning. As shown in [32,33,34] and our experiments in Section 6, incorporating structure into RPM problem solving could result in further performance improvement across different models.…”

Topological Relations Between Objects Are Categorically Coded

“…One set of models (Lovett & Forbus, 2017; Lovett & Forbus, 2011a; Lovett et al, 2009) builds on both communities, using categorical relations to compare images and solve problems. These models posit a vocabulary of categorical relations that could be used to solve a diverse set of visual comparison problems, inspired by behavioral studies (e.g., Huttenlocher, Hedges, & Duncan, 1991) and previous computational models.…”

Representation and Computation in Cognitive Models

“…Its performance on the Standard Ravens' test puts it in the 75th percentile, making it better than most adult Americans. Furthermore, it makes reaction-time predictions about human performance that have been confirmed in laboratory experiments (Lovett, Forbus, & Usher, 2010;Lovett & Forbus, 2017). 8…”