The installation “Mirage” uses principles from optics (astronomy/telescopes) and artificial neural network research to form a projection apparatus. Mirage generates a synthesised landscape based on its perception through a fluxgate magnetometer (Förster Sonde). A fluxgate magnetometer registers the magnetic field of the earth which is dependent on the suns activity and feeds it into a unsupervised learning algorithm for analyzation. At the same time the algorithm, that is based on the principle of a helmholz machine, “hallucinates” variations of the previously analyzed signal.
This variations are translated into a two dimensional matrix that physically transforms a thin mirror sheet by 48 muscle wire actors. The surface of the mirror sheet changes analog to the systems state. A thin laser line is directed on the mirror surface in a acute angle to generate a depth landscape like projection on the wall. Through the constant shifting signals the projection resembles a subliminal wandering through a landscape.
In 2013 Geoff Hinton, one of the leading researchers in the area of artificial neural networks and deep learning, joined Google to support them on various products that use AI and learning algorithms. He introduced back-propagation algorithms for training muli-layerd neural networks. One of his contributions to the field of unsupervised learing algorithms is the so called “Helmholtz Machine”, a machine which uses the principle of a wake-sleep-algorithm to consolidate its neural network. The alogrithm is trained during the wake phase by sensory input. In the sleep-phase it cuts-off its sensory input and feeds the network backwards with random patterns. On its input layer it generates versions of is previously perceived images of the world.
I am speculating that the computers in the enormous Google data-centers cut off their perception (search queries, user behaviour, speech regognition, image data) once a day and start to “sleep”. What do their “dreams” look like?
Ralf Baecker, 2014
Produced by LEAP Gallery, Berlin
Loading more stuff…
Hmm…it looks like things are taking a while to load. Try again?