VJC: Kendrick Kay
Currently, Nikhil, Robert Roodry from NYU and Jing Yang Zhou are organizing vision journal club on a bi-weekly basis (in-person at NYU and on zoom, 4pm Thursday). Zoom links are generated weekly for security issues. Please contact firstname.lastname@example.org to access the Zoom credentials.
This semester, we focus on looking at representations from perceptual, neuronal and theory perspectives. Here is a tentative schedule for the rest of the semester:
- 7/21/22 – Kendrick Kay, Assistant Professor at University of Minnesota
- Kendrick Kay’s talk will be at the Flatiron in the 4th Floor CCN Classroom
Title: Characterizing signal and noise in the Natural Scenes Dataset
Abstract: In this talk, I will first describe a recently released large-scale 7T fMRI dataset, termed the Natural Scenes Dataset (NSD). In NSD, whole-brain BOLD responses (1.8 mm, 1.6 s) were measured while each of 8 participants viewed 9,000-10,000 distinct color natural scenes (22,500-30,000 trials) over the course of 30-40 scan sessions. I will then present ongoing work on characterizing signal and noise in the NSD data. We develop a method and associated code toolbox for generative modeling of signal and noise distributions. This method has several applications: we show that the method can be used to estimate single-subject noise ceilings for representational similarity analysis, and can provide estimates of the dimensionality of signal and noise across visual cortex. One scientific finding is that the structure of spontaneous variability (noise) is well matched to the structure of the natural scene responses and this effect is enhanced at higher levels of the cortical visual hierarchy.