The Cortical Canvas
A theoretical ROAM paper framing spatiotemporal canvas allocation as an architectural analogue of cortical real estate, local/global connectivity, temporal hierarchy, and adaptive recurrent refinement.
This paper develops the theory side of ROAM: the claim that a shared spatiotemporal canvas can act like a uniform computational substrate whose layout, allocation, and connectivity encode the inductive bias for embodied intelligence.
The analogy is intentionally computational rather than biological. The paper compares bandwidth-proportional canvas allocation to cortical magnification, local dense plus sparse global attention to small-world cortical wiring, variable loop depth to adaptive metabolic allocation, and progressive attention sharpening to the feedforward-to-recurrent refinement cascade seen in cortical processing.
Its practical significance is that it turns a neuroscience metaphor into experiments: vary canvas real estate, test sparse topology thresholds, probe emergent specialization, measure adaptive loop allocation, and ask whether topology matters once the canvas is large enough for dense attention to stop being the default answer.
Neighborhood