will take place on Thursday, September 5th, 2024 from 15:00 to 16:00 hours in CBBM Building, Ground Floor, Seminar Room Loewi.
Host: Prof. Jonas Obleser
Institute of Psychology I
Abstract: The brain’s initial feedforward sweep is thought to perform a fixed mapping from sensory input to perceptual interpretation. Later feedback recurrence may iteratively refine this interpretation. But does the brain recruit feedback recurrence for image perception by default, and what type of objective function underpins feedback recurrence? In this talk, I will challenge the notion that feedback recurrence is not necessary when a stable bottom-up interpretation has already been established through prior familiarization with a stimulus. I will present evidence from a functional magnetic resonance imaging study demonstrating that feedback recurrence conveys more abstract and precise information for familiar than novel images, and introduce a computational model that explains how familiarity influences not the initiation but the content of feedback recurrence. Next, I will ask how different types of feedback recurrence shape the perception of challenging stimuli. I will present results from a modelling study that contrasts a network with generative feedback, trained to reconstruct noise-free images, against a network with discriminative feedback, trained to classify images. The differential image representations in these networks will be used to identify the type of feedback employed in the brain. Together, these studies offer insights into a dynamic neural architecture that balances between rapid and precise inference, advancing our understanding of the computational consequences of neural feedback.
