Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

The geometry of primary visual cortex representations is dynamically adapted to task performance

Leyla Roksan Caglar¹, Julien Corbo², O. Batuhan Erkat², Pierre-Olivier Polack; ¹Icahn School of Medicine at Mount Sinai, ²Rutgers University

Presenter: Leyla Roksan Caglar

Perceptual learning optimizes perception by reshaping sensory representations to enhance discrimination and generalization. Previous work has shown that learning a visual orientation discrimination task reshapes the population feature representations in the primary visual cortex (V1) via suppressive mechanisms. Although the computational importance of these changes has not yet been elucidated, it has been proposed that they optimize the geometry of the representation to be readout. Are these feature-encoding changes paired with changes to the representational geometry? To answer this, we investigated the relationship between V1 feature representation, behavioral performance, and neural manifold geometry in trained and naïve mice. Response dimensionality showed increases with task difficulty but was lower in trained animals, suggesting that successful learning reduces dimensionality. Based on manifold capacity, dimensionality, and radius, we further found that representational separability is a stronger predictor of individual behavioral performance. These results confirm that learning alters the geometric properties as early as early sensory representations, optimizing them for linear readout and improving perceptual decision-making.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF