This document defines a high level roadmap for sedna development.
The milestones defined in GitHub represent the most up-to-date plans.
- Support edge model and dataset management.
- Support incremental learning, with time trigger, sample size trigger, and precision-based trigger, and integrating hard sample discovering algorithm.
- Support collaborative training, integrating some common weight/gradient compression algorithm.
- Integrate some common multi-task migration algorithms to resolve the problem of low precision caused by small size samples.
- Integrate KubeFlow and ONNX into Sedna, to enable interoperability of edge models with diverse formats.
- Integrate typical AI frameworks into Sedna, include Tensorflow, Pytorch, PaddlePaddle and Mindspore etc.