Abstract
Categorical decisions based on noisy sensory information can be optimized by sequential sampling and integration. Cognitive psychologists have shown that human decision performance is limited not just by noise but by capacity, a constraint that emerges when competing sources of information are presented simultaneously. However, it remains unclear whether selective attention filters information at an early stage (during the processing of relevant sensory features) or at a late stage (during the integration of decision evidence). Here we recorded human EEG signals from 17 healthy subjects whilst they monitored two streams of Gabor patterns presented synchronously at 3 Hz to the left and right of fixation. At stimulation offset, subjects were probed to make a categorization judgment on one of the two streams. In the 'focused attention' condition, subjects were cued before stimulation onset as to which stream would be probed. In the 'divided attention' condition, which stream would be probed was only revealed after stimulation offset. We regressed human EEG signals against model-based variables representing the perceptual, decision and action-based information provided by each sample, and studied the stage(s) at which attention filtered information in the two conditions. Our results revealed two distinct stages at which attentional filtering occurs during sequential evidence integration. Diverting attention away from one stream by cueing it as irrelevant had a modest impact on the neural encoding of perceptual information, but precluded its conversion into decision information. By contrast, under divided attention, decision information from both streams was filtered only prior to integration, leading to a 'leaky' process that manifests itself as a bias to base decisions on the most recent evidence. The existence of distinct early and late filtering stages reconciles accounts of attentional selection that emphasize biased competition vs. a central bottleneck, and places important constraints on decision-theoretic models of perceptual categorization.
Meeting abstract presented at VSS 2014