Harnessing evolutionary computation to enable dynamically adaptive systems to manage uncertainty

This keynote talk and paper intend to motivate research projects that investigate novel ways to model, analyze, and mitigate uncertainty arising in three different aspects of the cyber-physical systems. First, uncertainty about the physical environment can lead to suboptimal, and sometimes catastrophic, results as the system tries to adapt to unanticipated or poorly-understood environmental conditions. Second, uncertainty in the cyber environment can have lead to unexpected and adverse effects, including not only performance impacts (load, traffic, etc.) but also potential threats or overt attacks. Finally, uncertainty can exist with the components themselves and how they interact upon reconfiguration, including unexpected and unwanted feature interactions. Each of these sources of uncertainty can potentially be identified at different stages, respectively run time, design time, and requirements, but their mitigation might be done at the same or a different stage. Based on the related literature and our preliminary investigations, we argue that the following three overarching techniques are essential and warrant further research to provide enabling technologies to address uncertainty at all three stages: model-based development, assurance, and dynamic adaptation. Furthermore, we posit that in order to go beyond incremental improvements to current software engineering techniques, we need to leverage, extend, and integrate techniques from other disciplines.