A recurrent vision transformer shows signatures of primate visual attention

Jonathan Morgan, Badr Albanna, James P. Herman · 2025

Read A recurrent vision transformer shows signatures of primate visual attention by Jonathan Morgan, Badr Albanna, James P. Herman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10955.txt.

Description

Jonathan Morgan, Badr Albanna, and James Herman built a recurrent vision transformer in two thousand twenty-five that shows signs of primate-like attention. Animal attention and AI self-attention are usually studied apart. Their model joins self-attention with a memory loop, so past and present inputs together steer where attention goes.

Trained only by sparse rewards on a cued change-detection task used in monkey studies, the model behaved like a primate. Cued items were caught faster and more accurately, scaling with cue reliability. Attention maps showed dynamic focus that lit up again before expected changes. Targeted damage caused shifts like those seen when primate attention areas are disturbed.

Catalog record

CREATORJonathan Morgan, Badr Albanna, James P. Herman
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-10955

Browse the full library · About the Tower of Babel