Federated Learning: Foundations and Applications

Learn the principles, workflows and challenges of federated learning, then put them into practice. Compare separate MNIST models trained on incomplete local data with a shared federated model, and investigate when collaboration improves generalization.

This 2.5-hour course combines the foundations of federated learning with a practical demonstration of learning across clients. The theory starts with a brief AI refresher and compares centralized, distributed and federated learning. It introduces the federated workflow, cross-device and cross-silo settings, horizontal and vertical learning, personalization, frameworks and training hyperparameters. Discussion covers non-IID data, communication cost, heterogeneous devices, security, privacy tools and evaluation at both global and client levels. In the practical session, participants split MNIST across three clients, each missing some digit classes. They train separate local models and examine their performance on a common evaluation set and on classes absent from local training. They then use the same client datasets in a FedAvg simulation, exchanging model updates while keeping raw training data local. Comparing the results shows when collaboration can improve generalization and where limitations remain. A follow-up assignment adapts the notebook to Fashion-MNIST with five IID clients. Federated learning is presented as a design choice with measurable trade-offs, not a guarantee of better accuracy or complete privacy.

Learning outcomes:

• Distinguish centralized, distributed and federated learning, explain a federated training round, and recognize the main federated learning settings.
  
• Run local-only and FedAvg training on three non-IID MNIST clients, and evaluate performance on classes missing from individual clients.
• Interpret global and client-level performance alongside communication cost and training efficiency, and explain the effects of non-IID data, system heterogeneity and privacy or security risks.
  

Target audience

Students, researchers and practitioners with basic machine learning experience who want to understand and implement learning across multiple data holders.

This webinar is open and free of charge for all participants from academia, industry, and public administration from EU and/or EuroHPC JU member countries.

Entry level & prerequisites

Basic Python, neural networks, training loops and model evaluation. Familiarity with PyTorch and Jupyter notebooks is helpful. A laptop with the course Python environment is required; clients will be simulated locally.

Agenda

For a detailed timetable and additional information, please see Agenda & Content in the left menu.

Speakers

Radu Prodan (University of Innsbruck)

Prof. Radu Prodan is a professor in the Department of Computer Science at the University of Innsbruck, where he leads the Edge AI research group. His research interests include distributed and parallel systems, cloud, edge and fog computing, social media networks, artificial intelligence and federated learning.

Turgud Valiyev (University of Innsbruck)

Turgud Valiyev is a research assistant and PhD student in Computer Science at the University of Innsbruck, working in the Edge AI research group led by Prof. Prodan. His interests include Edge AI, federated learning, machine learning and data science, with a focus on training and optimizing AI models for resource-constrained devices

Location

This course is offered as LIVE ONLINE WEBINAR via Zoom.

Registration

Registration is required, the link to the registration form can be found on the top of this page in the menu on the left. Please register with your official institutional email address to prove your affiliation.

You will get the Zoom link in the automatic confirmation by email (subject starting with "[Indico] Registration"), please check your Spam/Junk folders). We will also send 1–2 reminders before the webinar.

Please do not hesitate to contact us at training@ai-at.eu if you have any questions.

Modification, withdrawal & no-show policy

Your registration is binding. Please only register for the course if you are really going to attend.

You can update your registration data or withdraw your registration anytime before the registration form has been closed via the link "Manage my registration" which you can find at the bottom of your automatic email confirmation (subject starting with "[Indico] Registration").

Alternatively, or after the registration form has been closed, please inform us about your cancelation or any change in your registration data (especially your mobile-phone number) via email (training@ai-at.eu).

No-show policy: If you do not cancel and do not show up at the course you will be blacklisted and excluded from future training events.

Organizers

This webinar is organized by AI Factory Austria AI:AT.

Acknowledgements

AI Factory Austria AI:AT has received funding from the European High-Performance Computing Joint Undertaking (JU) under grant agreement No 101253078. The JU receives support from the Horizon Europe Programm of the European Union and Austria (BMIMI / FFG).