The Need for Semantic Interoperability in Distributed AI

As AI workloads move away from centralized cloud servers toward the edge-fog-cloud continuum, managing these distributed workflows becomes increasingly complex. The lack of a unified semantic framework prevents different infrastructure layers from communicating effectively, leading to fragmented deployments and inefficient resource allocation. The authors propose leveraging the Smart Applications REFerence (SAREF) ontology to bridge this gap, providing a standardized vocabulary that describes AI tasks, infrastructure capabilities, and data requirements across heterogeneous environments.

Leveraging SAREF for Workflow Orchestration

The proposed ontology extends the existing SAREF framework—originally designed for smart appliances—to accommodate the specific demands of distributed AI. By mapping AI workflow components (such as model inference, data preprocessing, and training tasks) to SAREF-based entities, the system allows for:

• Resource Discovery: Automatically identifying available compute resources at the edge, fog, or cloud level based on their semantic descriptions.
• Dynamic Task Mapping: Matching AI workload requirements (latency, memory, power) with the most appropriate infrastructure node.
• Interoperability: Enabling different platforms and vendors to exchange information about workflow states and resource availability without proprietary middleware.

By treating the edge-fog-cloud continuum as a single, semantically aware ecosystem, this approach simplifies the orchestration of complex AI pipelines that must balance local latency requirements with cloud-based processing power.