EdgeSmart: Configuring Complex AI Applications in Mobile Edge Computing

• Applicant: Professor Dr. Christian Becker
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 558100991

Edge computing applications are characterized by the use of services located at the interface between the wireless and the landline network. This enables functionalities, such as AI applications, that cannot be provided by resource-constrained mobile devices (energy, memory, CPU, etc.) and would incur a prohibitively high latency if executed in the cloud. In the past, edge computing research has focused on using or offloading a single service. In pervasive computing and in cloud settings, more complex application models have been investigated in order to configure multi-component applications with the resources that are available in the environment, based on application descriptions (contracts, architecture description languages). This typically involves optimization methods that map the configuration problem onto an optimization problem (CSP, ILP). In mobile edge computing, reconfigurations are typically required more frequently due to the inclusion of mobile resources. However, reapplying the computationally intensive optimization methods in response to every minor change in the environment is not very efficient. In relatively stable environments, learning methods can learn suitable configurations and find similar constellations efficiently. The objective of this project is therefore to develop an adaptive method that combines machine learning concepts with traditional constraint-based optimization methods and investigates which approach, depending on the situation (mainly with regard to the dynamics of mobile application scenarios), is more suitable and whether this can be automatically recognized and selected. To this end, the two configuration approaches are analyzed first separately and then conjointly. The aim is to develop an adaptive/hybrid method that – depending on the current situation – solves the configuration problem by applying the more suitable approach.

• DFG Programme: Research Grants