Tune in for the live stream on YouTube or Twitter.
Abstract: The production of archaeological knowledge is a pursuit inhibited by the quantity and quality of available information to be used in data analyses and interpretation. To combat the inherent shortcomings in our datasets, archaeologists have increasingly employed supervised machine learning algorithms to extrapolate patterns in known information to predict unknown information. While these pursuits have been productive, archaeologists are beginning to explore the research potential of unsupervised machine learning algorithms. A recent fluorescence in applying convolutional neural networks to museum collections offer a systematic way to classify ceramics, categorize the biological sex of human remains, and identify artifacts on the ground surface in remotely collected imagery using Unmanned Aerial Vehicles. While these applications are exciting in their ability to systematically produce data, they do not offer a means of using these data to further our understanding of the human experience in deep time. This presentation will provide an example of an artificial neural network application from the central Mesa Verde region in the northern US Southwest that pursues the production of knowledge using previously collected archaeological data. This novel application predicts periods of occupation and use of undated archaeological sites on an annual timescale based solely on surface ceramic assemblages. This is done by training a multi-label classification artificial neural network to quantify the relationship between proportions of ceramic assemblages and known dates of occupation at corresponding sites. The result is a demographic reconstruction of the central Mesa Verde region, from AD 450–1300, showing periods of population growth and decline at a finer temporal resolution than previously possible. The results from this model are then examined on the Mesa Verde North Escarpment to explore the human-scale of experience from AD 890–1300.