Feb. 17, 11:45, Wei Ji Ma. Optimality and Probabilistic Computation in Visual Categorization

Simon Kelly

Department of Biomedical Engineering

City College of New York

Electrophysiological signatures of decision formation in humans

Simple perceptual decisions are widely considered to comprise three stages: sensory encoding, decision formation and motor execution. By far the most mysterious has been the intermediate stage. Models drawing on the statistical framework of sequential sampling have invoked the concept of a decision variable (DV) to explain the accuracy and timing of decision reports. A DV represents a value that builds over time in proportion to the accrued evidence in favor of a given decision alternative, and determines action execution through a threshold-crossing criterion. There has been a surge in interest in this framework with the discovery of single-neuron signals that exhibit these very properties in association areas of monkeys. However, examining these processes in humans is far more difficult due to technical limitations. I will present the results of recent human electroencephalography (EEG) work in which we isolate and trace the dynamics of a supramodal decision signal that closely determines the timing and accuracy of actions in a way that transcends specific sensory or motor content. Our signal separation does not rely on decision model assumptions, nor on elaborate signal processing; rather, it falls out naturally due to very simple features of our task designs. I will demonstrate how we have established in our decision signal all of the dynamical properties that have been shown in monkey neurophysiology, and go on to describe new experiments in which we cover new ground on the mechanisms underlying speed-accuracy tradeoffs, incorporation of priors and evidence selection through top-down spatial attention.

Background Reading: Smith & Ratcliff (2004)