Discovering Contexts from Observed Human Performance
Abbreviated Journal Title
IEEE T. Hum.-Mach. Syst.
Context; clustering; context discovery; context-based reasoning; human; behavior representation; learning from observation; machine learning; TACTICAL HUMAN-BEHAVIOR; Computer Science, Artificial Intelligence; Computer Science, Cybernetics
This paper describes an investigation to determine the technical feasibility of discovering and identifying the various contexts experienced by a human performer (called an actor) solely from a trace of time-stamped values of variables. More specifically, the goal of this research was to discover the contexts that a human actor experienced, while performing a tactical task in a simulated environment, the sequence of these contexts and their temporal duration. We refer to this process as the contextualization of the performance trace. In the process of doing this, we devised a context discovery algorithm called context partitioning and clustering (COPAC). The relevant variables that were observed in the trace were selected a priori by a human. The output of the COPAC algorithm was qualitatively compared with manual (human) contextualization of the same traces. One possible use of such automated context discovery is to help build autonomous tactical agents capable of performing the same tasks as the human actor. As such, we also quantitatively compared the results of using the COPAC-derived contexts with those obtained with human-derived contextualization in building autonomous tactical agents. Test results are described and discussed.
Ieee Transactions on Human-Machine Systems
"Discovering Contexts from Observed Human Performance" (2013). Faculty Bibliography 2010s. 4771.