a The International Summer School in Federated Learning




Summer School Program


Morning Session:
  • Welcome sessions for MegaData: Federated Learning.
  • Machine Learning Recap.
Afternoon Session:
  • Introduction to Federated Machine learning

Morning Session:
Guest Talk by Daniel Beutel (FLOWER, UK)
  • Introduction to FL with FLOWER framework
Afternoon Session:
Guest Talk by Pan Heng (FLOWER, UK)
  • FLOWER practice session
  • Frameworks installation, configuration, and setup of FLOWER.

Morning Session:
  • Guest Talk by Javier Fernandez-Marques, FLLM with FLOWER.
Afternoon Session:
  • Guest Talk by Javier Fernandez-Marques, FLLM with FLOWER.
  • Team formation.

FedN full day session by Dr. Salman Toor, Co-founder of Scaleout, and Associate Professor at Uppsala University, Sweden
Title: Scalable and Secure Federated Learning: Architectures, Threats, and Practical Insights
Abstract: Scalable and Secure Federated Learning: Architectures, Threats, and Practical InsightsFederated Learning (FL) is transforming how machine learning models are trained across decentralized datasets, offering privacy-preserving solutions that align with modern data protection regulations and the growing need to keep data local. While FL enables collaborative learning without centralizing sensitive information, it also introduces significant challenges related to scalability, efficiency, and security, particularly vulnerabilities to adversarial attacks such as model poisoning. To address these risks, the session will introduce the attack landscape, explain its impact on FL, and present a novel defense approach using Conditional GANs (cGANs) to generate synthetic validation data on the server side, eliminating the need for external datasets while preserving privacy. This solution is designed to be scalable and easily integrated into FL frameworks like FEDn. In the second half of the session, participants will engage in a hands-on exercise where they will collaboratively train a machine learning model in a federated setting, introduce malicious clients, and observe the impact of attacks on model performance and integrity. This practical experience is intended to help participants better understand the unique security challenges in FL and how different types of attacks can compromise model robustness.

NVIDIA FLARE full day session by Dr. Zhijin Li, NVIDIA FLARE, USA
Title: Decentralized AI at Scale with NVIDIA FLARE
Abstract: NVIDIA FLARE (NVIDIA Federated Learning Application Runtime Environment) is a domain-agnostic, open-source, and extensible Python SDK for Federated Learning. It allows researchers and data scientists to adapt existing ML/DL workflow to a federated paradigm and enables platform developers to build a secure, privacy-preserving offering for a distributed multi-party collaboration. In this course, we will illustrate how quickly students can implement Federated applications using NVIDIA FLARE and move from simulation to production. Additionally, we will showcase some diverse applications implemented using NVIDIA FLARE across different industries. We will have materials for one full day. The students will be able to use our Deep Learning Institute platform for hands-on exercises using our hosted cloud compute resources.

Vantage6 full day session by Djura Smits, Netherlands eScience Center, Netherlands.
Title: An introduction to federated analysis using the vantage6 UI and python libraries.
Abstract: TBA.

Morning Session by Khoa Nguyen, Software Engineer at OpenMined
  • Title: Zero-Hassle Federated Learning for Diabetes Prediction with Syft-Flwr
  • Abstract: What’s the easiest way to leverage AI on sensitive patient data across multiple institutions—without ever moving a single record? In this three-hour, hands-on workshop, you’ll use Syft-Flwr, an open-source, flexible and offline-first protocol and tackle a real-world diabetes-prediction scenario. Alternating between Data Scientist (submitting, monitoring, tuning a training job) and Data Owner (provisioning SyftBox datasites, approving jobs), you’ll start with a local simulation across three hospital “nodes” before deploying it to Syft’s live network. Along the way, you’ll see how Syft’s open-network design accelerates discovery and enables true remote data science on distributed datasets. By the end, you’ll have built a privacy-preserving AI pipeline ready to scale across institutions, plus sample repositories and clear next steps for your own deployments.

Afternoon Session by Pedro Guijas Bravo, Co-founder of P2PFL, Spain.
  • Title: Decentralized Federated Learning: A Practical Overview and Demo of P2PFL
  • Abstract: Dive into Decentralized Federated Learning (DFL) with P2PFL, an open-source ecosystem for truly decentralized ML systems. This session will cover P2PFL's modular architecture. A hands-on demo will illustrate how to empower your machine learning models with P2PFL.

Morning Session by Dr. Ahmed Sayed, Associate Professor at Queen Mary University of London, UK.
  • Title: Robust Federated Learning for Network Monitoring
  • Abstract: In this tutorial, we will examine the use of federated learning (FL) to train a collaborative intrusion detection system and show the impact of presence of malicious clients on federated intrusion detection systems. We will use NSL-KDD, a commonly used (and sometimes criticized) dataset for training IDSs, and a simple neural network architecture to train our model. The IDS use case involves NIID data. We will also see how FL can be used to *share knowledge* between different institutions, but we assume that an attacker is sending malicious updates to negatively impact the system. We will see how the system can be made more robust to such attacks using a modified aggregation algorithm called FoolsGold.
Afternoon Session by Dr. Mehreen Tahir, Dublin City University, Ireland
  • Title: The art of Client Selection in Federated Learning: Multi-criteria client selection for efficient and secure FL
  • Abstract: Client selection plays a key role in improving the performance and security of Federated Learning (FL). This presentation explores a multi-criteria approach to client selection that balances efficiency, fairness, and security. We examine how factors such as data quality, system reliability, communication cost, and client behavior impact the learning process. By using adaptive strategies that combine these criteria, we can reduce training time, boost model accuracy, and defend against threats like data poisoning or unreliable clients. The goal is to present a practical framework for choosing the right clients at the right time to support robust and scalable FL systems.

All day:
  • Team working on projects coding, experimentation and writing report.
  • Preparing posters and demos.

All day:
  • Participant's projects and final presentations.
  • Summer school closing and photo session.