Abstract
What is the physiological basis of our adaptive perception and decision making? Recent studies suggest that the dynamic, task-dependent behaviors may be accounted by readout stage, rather than sensory encoding stages (Mante et al., 2013; Sasaki & Uka, 2009). However, from a computational viewpoint, dynamic modulation at the encoding stage can benefit the task performances when the encoding quality is confounded by its own system noise. Based on this idea, we questioned whether sensory cortex also show any adaptive dynamics, which depends on the task demands. We analyzed responses of color-selective neurons recorded in the macaque IT cortex, which change their activities depending on the task demands: discrimination or categorization (Koida & Komatsu, 2007). To clarify the functional meaning of the task-specific modulation, we focused on the stimulus representation at the level of neural population rather than at single neurons. We found that the task demands modulated the global pattern of activity distribution over the neural population. Importantly, the modulatory component had a stimulus-dependent dynamics after the stimulus presentation, suggesting that the modulation can be driven by the visual input, rather than static biases which is attributed to classical forms of attention. The pattern of dynamics was consistent with the categorical attraction, which is predicted by a recurrent network model that approximates the optimal probabilistic inference of stimulus dynamics based on a hidden-Markov model (Imai et al., 2011). Moreover, this modulatory effect was accompanied by selective increases in the task-relevant stimulus information conveyed by the neural population. The increases in encoding information indicates that the modulation of sensory representation is not a secondary effect, which simply echoes the dynamics in other area, but playing functional roles in the maximization of task-relevant encoding performances. These results suggest that the dynamic mechanism of task-dependent computation can include the sensory encoding stage.
Meeting abstract presented at VSS 2014