Characterizing neuronal population geometry with AI equation discovery

The visual cortex uses a high-dimensional code to represent sensory stimuli, but current understanding of single-cell tuning implies low dimensionality. We reconciled this discrepancy by developing an AI science system that used LLM-guided evolution to find a new parsimonious, interpretable equation for orientation tuning. The AI-discovered equation predicted tuning curves that are not smooth at their peaks, which we proved is required for high-dimensional coding. We used this equation to demonstrate the advantages of high-dimensional codes in a simulated hyperacuity task, and showed that similar coding occurs in head-direction cells. These results show that tuning smoothness has a key role in controlling information coding dimensionality, and moreover demonstrate how AI equation discovery can deliver concise models that accelerate scientific understanding in neuroscience and beyond.