Explaining object permanence with a deep recurrent neural network model of human cortical visual cognition

Objetivo

Visual cognition is our ability to recognize the things we see around us and make inferences about their meaning and
relationships. Deep convolutional neuronal network (CNN) models now achieve human-level performance on certain visual
recognition tasks and currently provide the most powerful models of human visual cognition. A hallmark step in the
development of human visual cognition is the acquisition of object permanence (OP). Object permanence is the ability to
continue to mentally represent an object that has disappeared from view – for example because it is hidden behind another
object. Current deep neural network models of vision lack this fundamental ability, limiting their power as models of human
visual cognition and as artificially intelligent systems. In this action, I will study the computational mechanisms necessary for
OP using a highly innovative approach that combines four elements: (1) a novel behavioral task that requires OP, (2)
development of a deep recurrent neural network models, (3) testing of both human participants and models at the task, and
(4) measurement of brain activity with functional magnetic resonance imaging (fMRI) during task performance. The OP task
involves viewing a scene of moving objects that occasionally become occluded behind other objects. Models will be trained
to represent objects continually, even as they vanish behind an occluder, and selected to match behavioral and cortical-layer-
resolved high-field fMRI data of human observers. The hosts, Prof Kriegeskorte at Columbia University and Prof Muckli
at University of Glasgow are world-leading experts on deep neural network models of vision and cortical-layer-resolved highfield
fMRI, respectively. The outcome of this action, a biologically plausible deep recurrent convolutional model that can
explain behavior and brain activity, will significantly enhance our understanding of the computational principles of visual
cognition, with implications also for AI technology.