Testing the Link Between Feature-Selective Attention and Choice-Probabilities in Primate V2

During perceptual decisions the activity of task-relevant sensory neurons is typically correlated with an animal's decision. Such decision-related activity, often quantified as "choice-probability", is thought to reflect feed-forward and feedback sources (Nienborg & Cumming, 2009). For a feature-discrimination task, this implies that the feedback is feature-selective. A recent computational account of choice-probabilities (Haefner et al., 2016) hypothesizes that the modulation of sensory neurons by feature-selective attention and the feedback source of choice-probabilities reflect the same mechanism. To test this hypothesis, we leveraged classical findings for feature-selective attention. When an animal attends to a stimulus feature, the responses of neurons selective for this feature are modulated (Treue & Martinez Trujillo, 1999). Critically, this modulation is observed globally, i.e. even when the attended, task-relevant stimulus is in the opposite hemifield to the receptive field of the modulated neuron. Together with the above hypothesis this finding makes a strong prediction: choice-probabilities should be observed globally, including for a task-irrelevant, ignored stimulus. To test this prediction we trained macaques on a disparity-discrimination task on one of two random-dot stereograms, each presented in one hemifield. Only one stimulus, always validly cued, was task-relevant and informative, and the hemifield in which it was presented was switched blockwise. Once the animals successfully ignored the task-irrelevant stimulus, we performed multichannel recordings from disparity-selective units in area V2. In support of the hypothesis, we find substantial choice-probabilities for the ignored stimulus (mean CP=0.59), slightly weaker than and correlated with those for the task-relevant stimulus (mean CP=0.62; r=0.41, p< 10-3, n=71). Importantly, this would not be expected in a feed-forward account, in which choice-probabilities reflect the read-out of the information used by the animal, or in feedback accounts, which differ from feature-selective attention. Instead, these results provide a novel, but predicted, link between feature-selective attention and decision-related activity in sensory neurons. Meeting abstract presented at VSS 2018