Abstract: This will be a 90 minute workshop that will walk through how to set up, run and deploy a federated learning project from scratch. The open source tools FEDn and STACKn will be used for model development, setting up a federated alliance, federated training and deploying the model to production. The code is open source and the tutorials will be available afterwards for all attendees.
Part 1 - Introduction
During this part we will present our solution to the data access problem. We will walk over what areas we will cover during the workshop and also look into the (real) but fabricated problem at hand we want to solve during this session.
We will furthermore also startup and briefly introduce the central components used.
Part 2 - Configuration
In this part we will work on setting up clients on distributed environments and going through some of the options that can be configured from the client perspective regarding execution environments and dataset allocation.
We will furthermore also load the federation with an untrained model and upload the compute package our clients run in order to do local training of the model.
We end this part by looking at metrics from model training and network behaviour when running several federated training cycles of the model.
Part 3 - Serving
In the third part we will have a look at deploying, accessing and serving model inferences to other services across an organization.
Part 4 - Iteration
In the fourth and last part we will work on iterating our own training out of the previously prepared example model. We will make tweaks to the model and create a new seed.
Finally, we will run another federation based on the new modified model and compute context.
A basic understanding of machine learning workflows, data pipelines and machine learning tools. The examples will be done in Python, but programming skills are not needed.
Bio: At Scaleout, we are solving the data access challenge in AI. We are developing a world leading solution for federated learning. In federated learning, you distribute the training of machine learning models to the data. You avoid collecting all data in one place.
Daniel is the CEO and co-founder of Scaleout and has a long background as an entrepreneur and leader in deep tech companies. He co-founded Scandinavia's first personal DNA-testing company in 2008, was CTO at a multinational growing medtech company for 7 years and then co-founded the first international accelerator for blockchain startups. As CTO and CEO, he has many years of experience in leading deep tech projects and taking them to market.