A graphical front end that can communicate with the kubeflow pipeline can save a lot of work for users who are not so familiar with the kubeflow environment. There are many laborious steps to take while constructing a pipeline. This is especially true for the Federated learning procedures which is subject to resource data format and number of clients. Not to mention the complex aggregating algorithms of the server.
A graphical front end that can communicate with the kubeflow pipeline can save a lot of work for users who are not so familiar with the kubeflow environment. There are many laborious steps to take while constructing a pipeline. This is especially true for the Federated learning procedures which is subject to resource data format and number of clients. Not to mention the complex aggregating algorithms of the server.
Currently, we use Node-red as frontend base for users to construct a kubeflow Federated Learning pipeline as bellow: https://github.com/sefgsefg/Federated-Learning-on-kubeflow-with-nodered/tree/main/FL_kubeflow_with_node-red This repo. can build a single Node-red node that can allow users to adjust the number of clients in the FL process. The control is then passed to kubeflow to construct a pipeline for execution.