search

xaynetwork / xaynet

Xaynet represents an agnostic Federated Machine Learning framework to build privacy-preserving AI applications.
https://www.xaynet.dev/
Apache License 2.0
200 stars 28 forks source link

Is it possible to have Federated Learning on Cloud-Edge? #563

Open Martiniann opened 4 years ago

Martiniann commented 4 years ago

Hi everyone,

Currently I am working on a school project about federated learning and came across your framework during exploratory analysis. My project should utilize federated learning in this manner - I have an aggregation server (let's say in a cloud). I want this server to provide model to my 2 Raspberry PIs. These two RPIs would then train the model on a local data for x epochs and provide the trained models/gradients back to the global server. On this server, the results would be federated averaged and new model would be sent to the PIs. Is such a workflow possible with your framework? If so, could you provide me a hint?

Thank you, Best regards

Robert-Steiner commented 4 years ago

Hi @Martiniann, in theory it is possible but unfortunately we don't provide the tools yet to easily support your setup.

We are currently working on building a client that runs on mobile phones. The client itself does not perform any machine learning, but only takes care of receiving / sending the global / local model and handling the PET protocol. To build your setup you will need to implement the machine learning part first and then integrate the client via FFI. However, this will require a lot of work. If you want to give it a try, let us know, we're happy to help.

In the near future we will provide a desktop client that will also support your use case.

prathapkumarbaratam commented 2 years ago

hi @Robert-Steiner , Iam currently working on project about federated learning and came across your framework during exploratory analysis. My project should utilize federated learning in this manner - I have an aggregation server (let's say in a cloud). I want this server to provide a model for my 2 Raspberry PIs. These two RPIs would then train the model on a local data for x epochs and provide the trained models/gradients back to the global server. On this server, the results would be federated averaged and a new model would be sent to the PIs. Is such a workflow possible with your framework? If so, could you provide me with a hint?any other examples using xaynet if possible?

Thank you,