"Repetition Suppression in Macaque Inferotemporal Cortex" and "Brain Network Connectivity and Graph Signal Processing"

CNBC Brain Bag
Center for the Neural Basis of Cognition (CNBC)

"Repetition Suppression in Macaque Inferotemporal Cortex" and "Brain Network Connectivity and Graph Signal Processing"

Nathaniel Williams and Mark Cheung
March 20, 2017 - 6:00pm
Mellon Social Room

Abstract:

The primate visual system, as studied at the level of neurons in inferotemporal cortex, is highly susceptible to functional modification by visual experience. One particularly robust effect of visual experience is repetition suppression. In repetition suppression, even a single exposure to a complex image reduces its ability to elicit a neuronal visual response upon subsequent presentation. There are two prevailing models of repetition-induced reductions of response strength in the visual system. These are fatigue based and correlation based. In a fatigue based model the selectivity of suppression can be no sharper than the selectivity of the initial response, since it requires that the cells or synapses active during the initial response become fatigued. If the mechanism is correlation-based, then the specificity of adaptation can be sharper than neuronal selectivity. In this talk I will present evidence that inferotemporal neurons are not selective for the conjunction of color and shape in an image but that repetition suppression is conjunction-specific. This finding supports the correlation based model and undercuts the fatigue based model of repetition suppression.

 

Abstract:

In neuroscience, a key challenge is to characterize networks underpinning cognition. In this talk, I will introduce several network modeling methods that can capture the dynamics of fMRI (and possibly other modalities), e.g., general correlation-based approaches such as inverse covariance estimation and lag-based approaches such as autoregressive modeling. Relatively new among these approaches, graph signal processing (GSP) is an emerging framework that merges algebraic and spectral graph theoretic concepts, allowing the extension of tools from classical signal processing to data indexed by graphs. An important analysis tool in GSP is the graph Fourier transform that can be found from the eigendecomposition of brain connectivity data, enabling graph signals to be decomposed into components that represent different modes of variability. I will present a particular GSP modeling approach, causal graph processes (CGP), to estimate effective brain connectivity by capturing dependencies among brain signals.