Zero-shot visual reasoning through probabilistic analogical mapping

• Published in: Nature communications Vol. 14; no. 1; p. 5144
• Main Authors: Webb, Taylor, Fu, Shuhao, Bihl, Trevor, Holyoak, Keith J., Lu, Hongjing
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 24.08.2023; Nature Publishing Group; Nature Portfolio
• Subjects: 631/477/2811; 639/705/117; Algorithms; Cognition & reasoning; Cognitive ability; Deep learning; Human performance; Humanities and Social Sciences; Mapping; multidisciplinary; Reasoning; Representations; Science; Science (multidisciplinary); Similarity; Training; Visual tasks
• ISSN: 2041-1723; 2041-1723
• DOI: 10.1038/s41467-023-40804-x

Human reasoning is grounded in an ability to identify highly abstract commonalities governing superficially dissimilar visual inputs. Recent efforts to develop algorithms with this capacity have largely focused on approaches that require extensive direct training on visual reasoning tasks, and yield limited generalization to problems with novel content. In contrast, a long tradition of research in cognitive science has focused on elucidating the computational principles underlying human analogical reasoning; however, this work has generally relied on manually constructed representations. Here we present visiPAM (visual Probabilistic Analogical Mapping), a model of visual reasoning that synthesizes these two approaches. VisiPAM employs learned representations derived directly from naturalistic visual inputs, coupled with a similarity-based mapping operation derived from cognitive theories of human reasoning. We show that without any direct training, visiPAM outperforms a state-of-the-art deep learning model on an analogical mapping task. In addition, visiPAM closely matches the pattern of human performance on a novel task involving mapping of 3D objects across disparate categories. Inspired by human analogical reasoning in cognitive science, the authors propose an approach combining deep learning systems with an analogical reasoning mechanism, to detect abstract similarity in real-world images without intensive training in reasoning tasks.