ePoster

PART–WHOLE REPRESENTATIONS IN LARGE VISION MODELS

Gabriel Mel

Centre de Recerca Matematica

FENS Forum 2026 (2026)
Barcelona, Spain
Board PS01-07AM-341

Presentation

Date TBA

Board: PS01-07AM-341

Poster preview

PART–WHOLE REPRESENTATIONS IN LARGE VISION MODELS poster preview

Event Information

Poster Board

PS01-07AM-341

Abstract

Foundation models provide a model system that (arguably) exhibits visual intelligence, is fully accessible to intervention and readout at all levels, and can communicate about its own internal states using natural language. Not only are such models an obvious experimental subject for visual neuroscience, they are also scientifically interesting objects of study in their own right. Drawing on ideas from visual psychology, systems neuroscience, and vision theory, we investigate how visual information is represented and transformed in multiple foundation models. Focusing on a particularly important aspects of visual function - part-whole structure - we find that models converge on broadly similar representation strategies, yielding a testable prediction for how part-whole structure in visual inputs is represented in the brain.

Recommended posters

Cookies

We use essential cookies to run the site. Analytics cookies are optional and help us improve World Wide. Learn more.