Deep Learning in Archaeology: Understanding the Composition of Ancient Mortars
Mortar is an essential part of construction, and has been used by builders for centuries. Ancient builders prepared them as members of changing communities of practice. But, to what degree did interactions among contemporaries lead to standardized mortars? Did builders learn from culturally different predecessors? In partnership with the Vanderbilt Data Science Institute, Dr. Markus Eberl, archaeologist and Vanderbilt University Associate Professor of Anthropology, aims to find the answers to these questions.
In the initial phase of this work, the project team will train a deep learning classifier to correctly identify particle types based on images of the individual particles. 1000 mortar samples – corresponding to ~1 billion particles with ~10 different images taken of each particle will be processed using a dynamic image particle analyzer. The samples will come from the ancient port city of Caesarea Maritima, which was built by Roman, Jewish, Byzantine, Abassid-Fatimid Muslim, and Crusader builders between 22 B.C.E. and 1265 C.E.
The interdisciplinary project team, composed of DSI staff data scientists, postdocs, DSI graduate students, and undergraduates will work with Dr. Eberl using machine learning algorithms to identify experimentally reproduced mortar constituents in the archaeological samples. The study’s foundational approach – dynamic image analysis, experimental archaeology, and machine learning – extends to other parts of the ancient Mediterranean – to others parts of the world – and even to other disciplines utilizing particles to understand composition of artifacts.