Gayness, Eigenforms, and Bodies

Éléments de Géométrie Biologique

Research Thread: p-adic Places and Visual Primitives

The Question

Can graphics primitives (the basic shapes CNNs learn to detect) be expressed as algebraic curves? And does the hierarchy of CNN layers correspond to something like places on those curves?

Starting Point: What Do Babies See First?

These are both degree-2 curves (conics), but the smile might be better modeled as a circular arc (constant curvature) rather than parabola (varying curvature). Perceptually simpler — one parameter.

What CNNs probably learn isn’t “parabola” or “circle” specifically, but:

The p-adic Connection

Khrennikov has been pushing p-adic cognitive models for decades. Key insight: neuron states as digits in p-adic expansion — each p-adic number represents a configuration of firing/non-firing neurons.

Recent development (Zúñiga-Galindo et al.):

Key paper: “p-adic cellular neural networks: Applications to image processing” — they determine all stationary states and can describe the dynamics almost completely.

The Open Question

Nobody seems to be connecting places on algebraic curves to visual feature hierarchies explicitly.

The existing work treats p-adic structure as convenient encoding for hierarchy. But:

The intuition: algebraic curves have Platonic existence, and what we perceive are shadows at various places. Graphics primitives might be curves, with the hierarchy of visual processing reflecting arithmetic structure.

People to Read

Next Steps