March 2026

Selective control of prefrontal neural timescales by parietal cortex

To understand how long-range connections between the prefrontal and parietal cortex shape the dynamics of attention, the authors investigate the frontal eye field of the macaque. The authors discover that the frontal eye field is divided into two distinct, specialized populations of neurons.

• Fast channel: Operating at a temporal resolution of 25ms, they serve as a fast channel for updating visual input and discriminating transient stimuli.
• Slow Channel: Operating at roughly 100ms, these neurons support sustained salient representations.

When the authors transiently deactivate the parietal cortex using implanted cryoloops, they found an asymmetric effect across channels. The fast channel was disproportionately affected, causing a global slowing of frontal eye field dynamics. Conceptually, the paper argues for a distributed circuit mechanism of attention control: parietal input is required both to maintain the speed of fast prefrontal updating and to tune the accuracy of slow prefrontal salience networks.

Regional specialization in the prefrontal cortex manifests in the reliability of task progression codes

Why might different brain regions encode the same information twice? The dorsomedial prefrontal cortex (dmPFC) and the orbitofrontal cortex (OFC) are classically thought to divide labor: the dmPFC handles action selection, while the OFC handles outcome prediction. Yet standard neural decoding shows that both regions represent the same thing. This apparent encoding degeneracy can be understood by changing the focus from what the brain encodes to when its dynamics become organized.

By simultaneously recording from large neural ensembles in the dmPFC and OFC of rats, the authors discovered a striking temporal dissociation:

• the action scaffold (dmPFC): the dmPFC provides a reliable representation of progression during specifically while traveling along the path
• the outcome scaffold (OFC): the OFC reliability peaks later, when arriving at the reward

When the dmPFC code degraded, the rats ran slower (an action-execution deficit). When the OFC code degraded, rats prematurely left the reward well before learning the outcome (a waiting expectation deficit).

This paper does fundamentally reframes how to analyze a region’s unique cognitive function: analyze its role as a temporal scaffold, not its static tuning.

Representational similarity modulates neural and behavioral signatures of novelty

How novel is a stimulus that is technically new, but shares features with something we’ve already seen? Standard computational models of novelty typically rely on strict “tally-based” mechanisms, which count how many times an exact stimulus was observed through discrete bins. This is brittle in naturalistic environments, where stimuli are continuous and may share structural overlap.

To solve this, the authors propose a similarity-based novelty model, where the stimuli are stored as a distribution over representational components. Experiencing a stimulus partially “familiarizes” the brain to unseen stimuli that share similar components. Novelty is then formally computed as the negative log of this similarity-generalized familiarity. Mathematically, this formulation is similar to Shannon surprise. However, whereas surprise is the negative log of expectation in the current context, novelty is the negative log of familiarity. The authors validate this new framework across two distinct domains:

• Sensory Processing (Mouse V1): When mice passively viewed repeated images with occasional novel images, a traditional tally-based model failed to explain the variance in their visual cortex response. The similarity-based model accurately predicted how novelty responses scaled with the similarity between the novel and familiar images.
• Spatial Exploration (Behavior): The authors treated physical locations in a maze as stimuli, encoded by place-field components across various spatial scales. A mouse’s novelty-driven exploration is best explained by a combination of exact state counts and similarity-based generalization.

Rapid concerted switching of the neural code in the inferotemporal cortex

Does the brain have a separate pathway for processing faces? A longstanding debate in neuroscience revolves around whether face-selective neurons in the inferotemporal (IT) cortex utilize a domain-general representation or a domain specific one. The authors measured neural activity in macaques while the animals viewed thousands of faces and non-facial objects. By embedding stimuli, they discovered a highly structured temporal dynamic.

• Early phase (Detection): Around 50-100ms after stimulus onset, the face-evoked neural axes align with the general object axes.
• Concereted Switch: At roughly 100ms, a rapid population event occurs for only facial images, lasting less than 20ms.
• Late Phase (Discrimination): Following this switch, the late code aligns with facial axes, specialized for fine face discrimination.

This offers a dynamic resolution to the domain vs general debate. The IT cortex begins with a general object code, and then dynamically switches to a domain-specialized one. It is suggested “Stimulus-gated switching” might be a more general representational mechanism in the cerebral cortex.