Modeling Focal Synaptic Degeneration and Neural Plasticity in Ventral Visual Cortex
Keywords: reorganization, recurrent connections, self-supervision, vision, stroke, deep artificial neural networks
TL;DR: Modeling focal synaptic degeneration and post-injury plasticity reveals physiological and anatomical reorganization in neurons and synapses, with spared recurrent connections and a contrastive learning retraining task improving recovery.
Abstract: Strokes affect a significant portion of the population and often result in secondary damage in the form of focal synaptic degeneration. When this occurs in the ventral visual cortex (VVC), it can lead to neurological deficits, including visual function loss. In this paper, we use the VVC as a framework in which to model focal synaptic degeneration and post-injury plasticity. We do so by progressively "injuring" synaptic connections in primate visual areas V1, V2, V4, and the inferior temporal cortex (IT), followed by continual retraining of the spared connections on real-world visual stimuli. We demonstrate that the functional signatures of carefully designed differential tasks can localize synaptic decay in the VVC. Initially, categorization performance deteriorates gradually, up to a critical threshold, beyond which there is a sharp drop. This slow decline in performance is marked by a reorganization in nearby neurons, where both visual function and the structure of receptive fields adapt to compensate for the damage. Spared recurrent connections significantly contribute to recovery. Furthermore, we find that the presence of teaching signals in the form of category labels during rehabilitation leads to improved categorization performance recovery.
Supplementary Material: zip
Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11112
Loading