2026-10-20

Animals exhibit behavioural flexibility by making reward-optimal decisions in unfamiliar situations. Computing value from minimal reward feedback requires leveraging known structure in the world. The Prefrontal Cortex (PFC) plays a pivotal role in both value computation and learning world structure, however mechanistic explanations for each---within the frameworks of Reinforcement Learning and Cognitive Maps---are largely distinct. We develop a mechanistic model of PFC that unifies value and structure through behavioural flexibility. In this model, value acts as a control signal that gates structured working memory representations of the current context in order to ultimately control behaviour. This is a significant departure from the conventional perspective of value, in which it is used for action readout. Critically, this model generalises to novel contexts without the need for additional synaptic plasticity. We demonstrate that recurrent neural networks (RNNs) meta-trained with either reinforcement learning or supervised learning recapitulate this exact model on various value-based tasks. These results not only provide a parsimonious explanation of both the value and schema frameworks of PFC function, but also propose a new mechanistic interpretation of the role of PFC in flexible decision making.

Longitudinal intravital imaging is essential for understanding how neural populations evolve across behavioral states, yet converting large-scale recordings into biological insight remains a fundamental challenge because neural signals are degraded by noise, non-rigid motion, and heterogeneous recording conditions. Here we introduce NeuroPilot, an agentic framework that transforms neural imaging analysis into a closed-loop, context-aware process, integrating a multimodal large language model with callable computational imaging modules to adaptively coordinate workflows from raw recordings to population-level readouts. Across heterogeneous imaging and experimental regimes, NeuroPilot restores motion-degraded recordings and supports consistent neural signal recovery, detection, and tracking in challenging deep-brain, freely moving, and longitudinal settings. These enable neuronal population tracking across spinal stimulation paradigms, deep-brain imaging, and continuous 24-hour mPFC recordings aligned with spontaneous sleep-wake behavior. By converting heterogeneous intravital recordings into stable population activity maps, NeuroPilot supports behavioral decoding and reveals structured neural dynamics across behavioral states and transitions. NeuroPilot establishes a scalable agentic paradigm for neural imaging analysis, expanding experimental access to the measurement, visualization, and interpretation of neural population activity.