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

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Behavioral relevance of high-dimensional neural representations

Chihye Han¹, Raj Magesh Gauthaman, Michael Bonner¹; ¹Johns Hopkins University

Presenter: Chihye Han

A common approach to understanding the organizing principles of neural representations has been to emphasize high-variance dimensions that correspond to interpretable features. Here, we investigate whether behavioral relevance is restricted to these interpretable dimensions or spans the entire spectrum of neural representations. Using fMRI data from the Natural Scenes Dataset, we tested whether humans could perceive coherent structure in image clusters formed along principal components of ventral visual stream responses, where explained variance decreases by orders of magnitude across principal-component ranks. In this initial study examining the first two decades of neural dimensions, we found that behavioral relevance extends throughout the entire range tested. These findings suggest that behaviorally relevant information in neural representations extends beyond the interpretable, high-variance dimensions emphasized in standard approaches and that comprehensive models of neural coding should account for the full range of dimensions.

Topic Area: Object Recognition & Visual Attention

Extended Abstract: Full Text PDF