Highly-patterned Found Objects

A project exploring online machine learning datasets as highly-patterned found objects by cataloging, documenting, and remixing them.

CONCEPT TEXT
During a web residency, I propose to explore online, publicly available machine learning datasets as patterned found objects. Faces, fingerprints, image captions, sea salinity, and other input data are typical, but so are more abstract patterns formed by neural network “weights” and thousand-dimensional vector spaces. Patterns in these datasets exist at several levels. They are highly-detailed conglomerations of minute datapoints; they evidence larger patterns in our priorities and scientific approach; and often they are seeded with moments of human language that defy the rigidness of the data (as in the case of a hand-written note I found in a 60-year catalog of solar data, buried in the US National Oceanic and Atmospheric Association FTP server, reading “Weather, suddenly + continually overcast”).

These datasets lie in wait across the web, in Github repositories, academic websites, and government FTP servers. Since an exhaustive investigation of these materials is not possible, this project will instead take an archaeological, morphological, and remix approach: cataloging, documenting, and manipulating the found materials in an exploratory way.

BIO
Jeff Thompson (b. 1982, Minneapolis/USA) is an artist, programmer, and educator based in the NYC area. His work explores collaboration with, empathy for, and the poetics of computers and technological systems. Through code, sculpture, sound, and performance, Thompson uses conceptual processes like remix, translation, and visualization to physicalize and give materiality to otherwise invisible processes. He is currently Assistant Professor and Program Director of Visual Art & Technology at the Stevens Institute of Technology, and co-founded the experimental curatorial project Drift Station.