Orthogonal neural encoding of targets and distractors supports multivariate cognitive control

Abstract

People can overcome a wide array of mental challenges by coordinating their neural information processing to align with their goals. Recent behavioral work has shown that people can independently control their attention across multiple features during perceptual decision-making, but the structure of the neural representations that enables this multivariate control remains mysterious. We hypothesized that the brain solves this complex coordination problem by orthogonalizing feature-specific representations of task demands and attentional priority, allowing the brain to independently monitor and adjust multiple streams of stimulus information. To test this hypothesis, we measured fMRI activity while participants performed a task designed to tag processing and control over feature-specific information that is task-relevant (targets) versus task-irrelevant (distractors). We then characterized the geometry of these neural representations using a novel multivariate analysis (Encoding Geometry Analysis), estimating where the encoding of different task features is correlated versus orthogonal. We identified feature-specific representations of task demands and attentional priority in the dorsal anterior cingulate cortex (dACC) and intraparietal sulcus (IPS), respectively, consistent with differential roles for these regions in monitoring versus directing information processing. Representations of attentional priority in IPS were fully mediated by the control requirements of the task, associated with behavioral performance, and depended on connectivity with nodes in the frontoparietal control network, suggesting that these representations serve a fundamental role in supporting attentional control. Together, these findings provide evidence for a neural geometry that can enable coordinated control over multiple sources of information.

Conflicts are parametrically encoded: initial evidence for a cognitive space view to reconcile the debate of domain-general and domain-specific cognitive control

“…The right lateralization of these regions is consistent with a previous finding (Li et al, 2017). The parametric encoding of conflict also mirrors prior research showing the parametric encoding of task demands (Dagher et al, 1999;Ritz & Shenhav, 2023). The scaling of brain activities based on conflict difference is potentially important to the representational organization of different types of conflicts.…”

Distractor-specific control adaptation in multidimensional environments

“…In our multidimensional task, this predicts generalization of adaptation across dimensions, at odds with our data. However, complementary mechanisms may operate through suppression of single interfering stimulus features, their orthogonalization (Flesch et al, 2022;Ritz & Shenhav, 2024), or through an increase in dimension separability (Badre et al, 2021). In conclusion, hypotheses based on 'local representations' alleviate the need for a high degree of flexibility in connectivity between ACC and posterior regions, but it comes at the expense of an (inefficient) duplication of information from more sensory areas to medial frontal cortex.…”

Task structure tailors the geometry of neural representations in human lateral prefrontal cortex

“…25 . More generally, orthogonal coding has been consistently observed in lPFC in task settings with potential for interference, such as between the coding of targets and distractors 44 , and between previous and current task states 31 , suggesting a specific role for lPFC in encoding orthogonal subspaces in the service of cognitive control 29 .…”