Formal research on perception, and its relation to the studio work.
Space Between Minds: A Geometry of Perception and the Representation That Could See
What if the space between two minds had a geometry?
Taminelli, R. (2026). Space Between Minds: A Geometry of Perception and the Representation That Could See. Zenodo.
https://doi.org/10.5281/zenodo.20758940
ORCID: 0009-0006-2026-4574
Code: github.com/robertotaminelli/SPIM
Fisher-Rao metric · statistical manifold · THINGS database · odd-one-out similarity · Riemannian geometry of observers
What the paper does
Two observers exposed to the same world do not organise it the same way. This paper asks whether that difference can be measured.
The Space of Perceptual Internal Models applies the Fisher-Rao metric across observers rather than within them. Each observer’s internal generative model is estimated as a probability distribution over perceptual objects, derived from odd-one-out similarity judgments, and placed as a point in a statistical manifold. The distance between two points measures the statistical distinguishability of their internal models: how differently two minds organise a shared world.
The instrument is built on the THINGS database (Hebart et al., 2020), using 1500 observers divided into three consecutive batches of 500. It produces stable, replicable structure: the same core-periphery density gradients appear in all three batches. A noise-null control, running the identical pipeline on synthetic data, produces flat geometry. The
instrument does not invent structure where there is none.
It also states its own limit. At current sampling density, the geometry is reproducible from sampling assignment alone, which means the design cannot yet separate perceptual signal from sampling structure. The paper identifies precisely the condition that would lift this: a partially orthogonal anchored design, in which all observers co-observe a shared anchor set while otherwise receiving non-overlapping objects.
The result is not a model of perception. It is an instrument for measuring the distance between minds, and a precise account of what it would take to sharpen its resolution.
Painting and Code
Every work is a communicative act, like every word, every glance, every gesture. What it carries reaches other minds. What produced it does not, and we are not trained to notice the difference.
The perceptual act through which a mind takes in the world and gives it back is available only to whoever lives it. The work carries the residue of a single perception. Another mind reads it through its own. The distance between the two is what the paper sets out to measure.
Further papers in development on the same research.
Lighted Angles
Thesis, 2021. English translation and Zenodo deposit in preparation.
A commentary on the Munker-White illusion. A square (and squares) of uniform luminance placed over four lateral gradients, with a thin border carrying the gradient color at the edges.
The argument: the visual system constructs perception in layers. It resolves the structural elements first (the bounding lines and the angles) and only then adjusts the color reading of the regions contained within those structures. Each layer is conditioned by the one before it.
The Lighted Angles effect makes this architecture visible: the perceived curvature at the corners is not in the signal. It is produced by the system’s own prior constructions reading back into the current input.
Perception does not receive. It builds, and then rebuilds, layer by layer.
English translation and Zenodo deposit in preparation.