Stable readout of observed actions from format-dependent activity of monkey’s anterior intraparietal neurons

Significance

The anterior intraparietal area (AIP) is a crucial hub in the observed manipulative action (OMA) network of primates. While macaques observe manipulative action videos, their AIP neuronal activity robustly encodes first the viewpoint from which the action is observed, then the actor’s body posture, and finally the observed-action identity. Despite the lack of fully invariant OMA-selective single neurons, OMA exemplars could be decoded accurately from the activity of a set of units that maintain stable OMA selectivity despite rescaling their firing rate across formats. We propose that by integrating signals multiplicatively about others’ action and their visual format, the AIP can provide a stable readout of OMA identity at the population level.

Abstract

Humans accurately identify observed actions despite large dynamic changes in their retinal images and a variety of visual presentation formats. A large network of brain regions in primates participates in the processing of others’ actions, with the anterior intraparietal area (AIP) playing a major role in routing information about observed manipulative actions (OMAs) to the other nodes of the network. This study investigated whether the AIP also contributes to invariant coding of OMAs across different visual formats. We recorded AIP neuronal activity from two macaques while they observed videos portraying seven manipulative actions (drag, drop, grasp, push, roll, rotate, squeeze) in four visual formats. Each format resulted from the combination of two actor’s body postures (standing, sitting) and two viewpoints (lateral, frontal). Out of 297 recorded units, 38% were OMA-selective in at least one format. Robust population code for viewpoint and actor’s body posture emerged shortly after stimulus presentation, followed by OMA selectivity. Although we found no fully invariant OMA-selective neuron, we discovered a population code that allowed us to classify action exemplars irrespective of the visual format. This code depends on a multiplicative mixing of signals about OMA identity and visual format, particularly evidenced by a set of units maintaining a relatively stable OMA selectivity across formats despite considerable rescaling of their firing rate depending on the visual specificities of each format. These findings suggest that the AIP integrates format-dependent information and the visual features of others’ actions, leading to a stable readout of observed manipulative action identity.