Capso: A Multi-Objective Cultural Algorithm System To Predict Locations Of Ancient Sites

Samuel Dustin Stanley, Wayne State University






August 2019

Advisor: Dr. Robert Reynolds

Major: Computer Science

Degree: Doctor of Philosophy

The recent archaeological discovery by Dr. John O’Shea at University of Michigan of prehistoric caribou remains and Paleo-Indian structures underneath the Great Lakes has opened up an opportunity for Computer Scientists to develop dynamic systems modelling these ancient caribou routes and hunter-gatherer settlement systems as well as the prehistoric environments that they existed in. The Wayne State University Cultural Algorithm team has been interested assisting Dr. O’Shea’s archaeological team by predicting new structures in the Alpena-Amberley Ridge Region.

To further this end, we developed a rule-based expert prediction system to work with our team’s dynamic model of the Paleolithic environment. In order to evolve the rules and thresholds within this expert system, we developed a Pareto-based multi-objective optimizer called CAPSO, which stands for Cultural Algorithm Particle Swarm Optimizer. CAPSO is fully parallelized and is able to work with modern multicore CPU architecture, which enables CAPSO to handle “big data” problems such as this one.

The crux of our methodology is to set up a biobjective problem with the objectives being locations predicted by the expert system (minimize) vs. training set occupational structures within those predicted locations (maximize). The first of these quantities plays the role of “cost” while the second plays the role of “benefit”. Four separate such biobjective problems are created, one for each of the four relevant occupational structure types (hunting blinds, drive lines, caches, and logistical camps). For each of these problems, when CAPSO tunes the system’s rules and thresholds, it changes which locations are predicted and hence also which structures are flagged. By repeatedly tuning the rules and thresholds, CAPSO creates a Pareto Front of locations predicted vs. structures predicted for each of the four occupational structure types.

Statistical analysis of these Pareto Fronts reveals that as the number of structures predicted (benefit) increases linearly, the number of locations predicted (cost) increases exponentially. This pattern is referred to in the dissertation as the Accelerating Cost Hypothesis (ACH). The ACH statistically holds for all four structure types, and is the result of imperfect information.