Population codes for context-dependent decision-making

Context-dependent decision-making enables flexible behavior by allowing identical sensory inputs to guide different actions depending on memory, rules, or goals. Recent advances in large-scale neural recordings have shifted the focus from single-neuron tuning to population-level representations, revealing principles by which neural populations support such flexibility. Here, we review evidence of how context-dependent decisions are implemented by population coding mechanisms, including nonlinear mixed selectivity, task-dependent population geometry, shared representational subspaces, and structured across-neuron correlations. Nonlinear mixed selectivity expands representational dimensionality, allowing downstream readout of arbitrary combinations of task variables. Learning reshapes population geometry, and it may balance flexibility and generalization by promoting the reuse of shared representations when task components overlap. Structured correlations between neurons that share a projection target enhance transmission of context-dependent information to downstream circuits. These population-level coding mechanisms provide a conceptual framework for understanding how neural circuits integrate sensory and contextual information to guide behavior.