SPACE4AI-D
A Design-time Tool for AI applications Resource Selection in Computing Continua
The SPACE4AI-D tool tackles the design space exploration of AI application component placement and resource selection challenge at design time to minimise the execution cost of the AI application while providing performance guarantees. It implements several heuristic algorithms such as random greedy, simulated annealing, genetic algorithms, local search, and tabu search.
First tool that considers AI applications with different candidate deployments which include different DNN partitions for each component and jointly considers resource contention by analysing different resource candidates. Its purpose is to select the optimal deployment and target resources to minimise the execution cost while guaranteeing response time performance constraints.
SPACE4AI-D, within AI-SPRINT, explores design alternatives to minimise costs while coping with technology constraints, performance, and privacy requirements. It identifies optimal component placement and resources of the computing continuum with different computational layers and heterogeneous nodes.