kubeedge/sedna

Sedna is an edge-cloud synergy AI toolkit that enables collaborative training and inference across distributed edge and cloud environments.

Sedna addresses the challenge of deploying machine learning workloads across heterogeneous edge-cloud infrastructure by providing frameworks for joint inference, incremental learning, federated learning, and lifelong learning. It works by leveraging KubeEdge's orchestration, messaging, and service discovery capabilities to coordinate AI tasks between edge devices and cloud servers. The toolkit allows developers to add edge-cloud synergy capabilities to existing training and inference scripts with minimal modification, supporting popular frameworks including TensorFlow, PyTorch, PaddlePaddle, and MindSpore.

Teams should adopt Sedna when they need to reduce inference latency and bandwidth costs by offloading difficult tasks to the cloud while keeping simple inference on edge devices, or when they must train models on distributed edge data without centralizing raw data due to privacy or scale constraints. It suits scenarios with small samples and non-independent-identically-distributed data on edge nodes, as well as applications requiring continuous learning across heterogeneous data distributions. The toolkit provides dataset and model management across edge-cloud boundaries, preset algorithms for hard sample discovery and parameter aggregation, and extensible interfaces for integrating custom algorithms.

The project maintains active continuous integration workflows and demonstrates ongoing development through regular updates to its codebase. Development activity shows consistent engagement with the repository through code contributions and maintenance of the core framework components.