Handling Seasonality using Metacognition

摘要

This paper summarizes a work in progress in the area of the metacognitive loop (MCL). The objective of MCL is to provide a design approach supported by software to extend an intelligent system's ability to cope with perturbations. A perturbation is any deviation from optimal performance for the system. Many MCL implementations exist, each increasing in sophistication. This paper describes an approach to produce the next implementation of MCL, which we call the General Purpose Metacognition Engine (GPME). The GPME evolves the functionality of the current implementation developed at the University of Maryland, MCL2, in particular, to handle seasonality. Seasonality is a periodic or cyclic variation in conditions that causes agents to re-learn when the length of the seasonal cycle exceeds their ability to detect the cycle.