Integration of Hybrid AI With Cloud Edge and High Performance Computing

The extreme proliferation of intelligent applications on the spectrum of cyber-physical systems has revealed structural dis-alignment between inertial cloud-centric AI implementations and the dynamic nature of the real-time, heterogeneous workload. Present hybrid architectures are largely based on rigid task allocation or rule-based coordination, and do not provide flexibility when workload intensity and mixed computational requirements vary. The chapter suggests a hybrid AI system that closely combines edge intelligence, cloud-native coordination, and high-performance computing via learning-based coordination. The framework allows the migration of the intelligence to be adapted according to the workload nature, state of the system, and performance feedback as opposed to binding the models to fixed layers. High-performance resources are not considered passive accelerators, but instead viewed as active decision partners, which allows them to be optimized on a scale and make inferences in real-time.

Перевод пока недоступен