Through the Eyes of the Machine

Exploring Historical Photo Collections with Convolutional Neural Networks

Abstract

The article deploys Convolutional Neural Networks to cluster historical photo collections by visual similarity to aid exploration and mediation. It examines potentials and challenges of human-machine collaboration by juxtaposing human and machine ways of seeing. Clustering 48,000 negatives from the collection Ernst Brunner the analysis reveals how sociotechnical imaginaries in infrastructure act as an epistemological Trojan horse and emphasizes the need for thematic data sets to utilize machine-learning approaches for visual data analysis.