The presentation discusses how KubeEdge Sedna, a cloud-native edge machine learning suite, is combined with TinyMS, a high-level API toolkit for MindSpore deep learning framework, to enable incremental learning at the satellite to accomplish tasks like remote sensing and earth observing.
- KubeEdge has brought cloud-native to space with several small research satellites equipped with edge computing and AI.
- KubeEdge Sedna and TinyMS are combined to enable incremental learning at the satellite for tasks like remote sensing and earth observing.
- The presentation provides a deep dive into how KubeEdge Sedna and TinyMS work together to accomplish incremental deep learning for satellites.
- The use of KubeEdge Sedna and TinyMS allows for the optimization of vast amounts of data using artificial intelligence and cloud-native technology.
- The presentation includes an anecdote about how KubeEdge Sedna and TinyMS are used to detect farmland areas using satellite data.
The presentation provides an example of how KubeEdge Sedna and TinyMS are used to detect farmland areas using satellite data. The image detection model is trained using MySport with tremendous amounts of data, and the pre-deployed model is served with KubeEdge Yamna via TinyMS. The model can detect farmland areas by itself, but in some cases, low precision models can result in inaccurate detections. When hard samples are discovered, the local controller compresses and sends the datasets to the ground using a high precision image detection model for inference. These hard samples can also be used to train a new model and improve precision. The global manager performs incremental learning tasks using MySport to train the model, which is fine-tuned and retrained by MySport. KubeEdge Sedna's semantics are integrated as an integral part of the AI framework.