Fold

A network bending the plane until two tangled things come apart.

Two spirals are wound through one another so that no straight line, and no gentle curve, can tell them apart. A network is seeded knowing nothing and given the one task of separating them — and the only way is to bend the plane itself. Fold shows that bending: the field of the network's belief warping in real time, growing arms that reach around each spiral until every point has come to rest on its own side. When the tangle is solved the fold holds, clean, for a while. Then it dissolves, a new tangle is drawn, and it begins again. No two are alike. The work is built to run for hours, and does not repeat.

Image

The whole frame is the plane the network is learning. The colour at every point is its answer there — warm for one class, cool for the other, dark where it is certain, and a luminous seam along the boundary between: the fold. At the outset the field is an undifferentiated glow; as the network learns, the seam bends and folds into the shape of the tangle. The points are the truth it is fitting; a point still on the wrong side glimmers until the fold catches it.

Sound

Sound is continuous and made in the moment. Each point the fold catches rings a single note, drawn from where it sits in the tangle, so the melody sweeps around as the boundary closes. Many notes while the plane is still tangled, then fewer as it comes apart. Nothing is sequenced and nothing is recorded. What is heard is a function of what the network is folding, right now.

The artist

Joshua Borsman is an artist working with sound, light, and real-time systems. His generative works give form to processes that ordinarily pass unseen — the orbits overhead, the weather of a coast, the machinery beneath the internet — and render them as continuous, non-repeating fields of image and sound. Fold is the second work in Hidden Layers, a series on the unseen interior of machine learning — the place where a system comes to know.

Specification

Series: Hidden Layers, II
Form: Generative real-time installation
Process: In-browser neural network · live gradient descent
Image: Real-time canvas
Sound: Continuous; rooted in A
Edition: Browser; all platforms

← Return to the work