1. Discovering Tactical Memory From Observed Human Performance in Machine Learning
- Author
-
Avelino J. Gonzalez and Josiah Wong
- Subjects
Training set ,Computer Networks and Communications ,Computer science ,business.industry ,Human Factors and Ergonomics ,Machine learning ,computer.software_genre ,Computer Science Applications ,Task (project management) ,Human-Computer Interaction ,Memory management ,Artificial Intelligence ,Control and Systems Engineering ,Lawn mower ,Signal Processing ,Task analysis ,Memory modeling ,Artificial intelligence ,Representation (mathematics) ,Set (psychology) ,business ,computer - Abstract
This article describes an investigation for composing a representation of significant past events that are retained in memory by an observed human actor. These memories influence that actor's performance of a task or making a decision. More specifically, we seek to infer which aspects of the environment and which events have significant effect on an observed actor's future decisions. We introduce a new memory modeling algorithm, memory composition learning, which processes traces of an observed actor's performance, and from these, composes a set of memory features that describe important events in his/her memory that affected these actions. These memory features are subsequently used to produce memory-enhanced traces that can be used by machine learning algorithms to learn memory-influenced behaviors. We implemented a prototype of our approach and evaluated it in two simulated domains, one with synthetic memory-influenced vacuum cleaner agents and one involving human subjects controlling a lawn mower that required memory-influenced behaviors. Results show that our approach is able to discover intuitive representations of tactical memory from observed behavior in both domains, and that these memory representations contributed to improved machine learning of human behavior.
- Published
- 2021
- Full Text
- View/download PDF