Marieke Mur (AP Math Colloquium)

Marieke Mur - The Brain and Mind Institute

Title: Dynamic population codes in the primate visual system

Abstract: The primate visual system is generally modeled as a hierarchical feedforward system. The system transforms visual images (input) into a semantic representation (output) through a cascade of linear and nonlinear transformations. The feedforward framework is quite successful at modeling how the visual system infers meaning from images; however, it ignores temporal dynamics. Temporal dynamics may be highly relevant for object recognition. I will present analyses of cell recordings from macaque inferior temporal (IT) cortex, the output stage of the primate visual system. Neural activity was recorded while macaques viewed object images from a range of categories (e.g. fruits, faces). The analyses examine how neural population coding of object images evolves over time. Results indicate that both within- and between-category distinctions arise early (80 ms after stimulus onset). The within-category distinctions can be explained by low-level features computed from the input image, whereas the between-category distinctions cannot, especially at later time points after stimulus onset (>125 ms). These results suggest that the IT code reflects both visual similarity and category distinctions and that the latter may require recurrent processing to emerge.