From Circuits to Behavior: Modeling Flexible Context-Driven Sensory Processing

Visual representations become progressively more abstract along the cortical hierarchy. These abstract representations define notions like objects and shapes but at the cost of spatial specificity. By contrast, low-level regions represent spatially local but simple input features. How do spatially non-specific representations of abstract concepts in high-level areas flexibly modulate the low-level sensory representations in appropriate ways to guide context-driven and goal-directed behaviors across various tasks? We present a biologically motivated and trainable neural network model of dynamics in the visual pathway, incorporating lateral recurrent and feedforward synaptic connections, distinct excitatory and inhibitory neuronal types, and long-range top-down feedback from high-level areas conceptualized as low-rank modulations of the input-driven sensory responses. We train this model to solve a difficult visual cue-delay-search task. On each trial, the model learns to modulate its perceptual responses based on the visual features of the cue, to solve the task. Notably, it outperforms state-of-the-art DNN vision and LLM models with orders of magnitude fewer parameters. In silico electrophysiology of trained models reveals key neural tuning properties and excitation-inhibition dynamics. The model exhibits emergent excitation-inhibition balance, macroscopic gradients in neural timescales, and feature selectivities across cortical layers. Inhibition-lesioned versions of the model show diminished stability and expressivity. We fine-tune the same model on two classic psychophysics attention tasks (“feature” searches). We find that reaction time measures derived from model dynamics closely replicate known behavioral signatures from the human psychophysics literature. This work represents a powerful tool for studying recurrent and contextual sensory processing and lays the groundwork for future research on the neural mechanisms of cortical feedback and attention.