A taxonomy of artificial intelligence approaches for adaptive distributed real-time embedded systems
Jeremy Davis, Joe Hoffert, Erik Vanlandingham · 2016
Distributed real-time embedded (DRE) software systems such as are used to manage critical large-scale infrastructure are important systems to target for increased functionality and resiliency. DRE systems that can adapt to changes in the environment and/or changes in available resources are more robust to unexpected changes and extend both the systems' utility and lifespan. Artificial intelligence techniques are used for adaptation of software systems in general. However, they must meet certain requirements to be appropriate for use with DRE systems. Any artificial intelligence (AI) technique used in an adaptive DRE system should produce consistent results across a distributed system, operate in bounded time, make decisions autonomously, and gracefully handle and learn from previously unencountered environments. This paper surveys a variety of AI techniques, providing a brief overview of each method, evaluating how each technique fits the requirements of an adaptive DRE system, and recognizing the gaps in each technique's ability to meet all of these requirements. Our results show that there is not one single AI technique in our survey that is a perfect fit for adaptive DRE systems, although some techniques address more of these requirements than others.