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

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

Sample, Don't Assume: Unconstrained Receptive Field Estimation

Niklas Müller¹, Paolo Papale², Iris Groen¹; ¹University of Amsterdam, ²Netherlands Institute for Neuroscience

Presenter: Niklas Müller

The spatial tuning of (populations of) neurons, as reflected in their (population) receptive field (pRF), is one of the most fundamental properties determining neural responses in visual cortex. pRF geometry is typically modeled as a 2D isotropic Gaussian, effectively assuming the pRF samples a circular 'aperture' in the visual field. However, it has been found that using a more complex geometry can improve neural predictions. Thus, it remains unclear what assumptions to make about the geometry of pRFs. Here, we show that removing any geometrical assumptions, and instead estimating pRFs in a fully data-driven way, leads to significant improvements in neural predictions. We combine linear encoding models with random sampling of pixels from feature maps of convolution deep neural networks to estimate unconstrained pRFs from monkey multi-unit electrophysiology recordings. Our new method not only improves neural predictions but also allows for both quantitative pRF mapping (parameter estimation) and qualitative inspection of the pRF geometry from the obtained pixel importance maps.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF