Brown University

Evaluating and Improving Models of Visual Perceptual Learning

Description

Abstract:
Abstract of Evaluating and Improving Models of Visual Perceptual Learning by Tyler Barnes-Diana, Ph.D., Brown University, May 2024. An evaluative framework for investigating computational models of Visual Perceptual Learning (VPL) is described that focuses on qualitative replication of characteristics of human VPL across tasks. Characteristics include learning, specificity, and stimulus ordering effects. Three feedforward network models of VPL, two neural networks of varying architectures and the Integrated Reweighting model, are evaluated on this framework. On a contrast discrimination task all three models failed to replicate specificity to reference contrast as well as replicate human behavior under “roving” conditions, whereby with stimulus ordering conditions of interleaved adaptive staircases humans fail to learn. Failure to replicate human behavior under roving conditions may reflect a lack of memory or recursion system in the models presented here. Model updates are suggested that may improve model performance on the evaluative framework described. Suggested model updates, informed by both behavioral work as well as modeling results, include 1) a multistage model of VPL that considers the multiple stages of the learning process that have been observed in behavioral work as well as 2) recursive models of VPL that may provide a system to capture temporal relationships in stimuli that may drive stimulus ordering effects in humans. Preliminary results of recursive models are shown that demonstrate learning on a contrast discrimination task.
Notes:
Thesis (Ph. D.)--Brown University, 2022

Citation

Barnes-Diana, Tyler, "Evaluating and Improving Models of Visual Perceptual Learning" (2022). Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations. Brown Digital Repository. Brown University Library. https://repository.library.brown.edu/studio/item/bdr:qp4wezuq/

Relations

Collection: