SC2024 Workshop and Technical program presentations & involvement by UvA group members.
- The INDIS 2024:
The 11th Annual International Workshop on Innovating the Network for
Data Intensive Science - INDIS, Monday Nov 18, 9:00 - 17:30 MST in room
B305
- Session Chairs
- Akbar Kara, Ciena Corporation, SCinet
- Anees Al-Najjar, Oak Ridge National Laboratory (ORNL), SCinet
- Nik Sultana, Illinois Institute of Technology, SCinet
- Cees de Laat, University of Amsterdam, SCinet
- Description
- The needs of science networks are rapidly evolving.
High-volume data distribution in High Energy Physics, Astronomy, and
Light-Sources is seeing a new wave of Artificial Intelligence
(AI)-dominated research. Not only AI is driving applications on the
networks, but AI now also gets applied in harnessing the complexity of
the networks. Potential applications of Quantum networks and deployed
network testbeds, possibly connected with regional supercomputers, can
serve the new science demands. The proposed INDIS workshop encourages
high-end research and state-of-the-practice papers that address one or
more of these networking needs, and developments that are essential in
the cyberinfrastructure for the scientific discovery process. The
workshop also serves as a platform for participants in Network Research
Exhibitions, Experimental Networks of the Future, and SCinet to submit
and present papers on their latest innovations, designs, and solutions,
and also to showcase the next generation of networking challenges and
solutions for HPC.
- Program
- INDIS presentation: "Secure Collaborative Model Training with Dynamic Federated Learning in Multi-Domain Environments", Anestis Dalgkitsis,
Alexandros Koufakis, Jorrit Stutterheim, Aleandro Mifsud, Priyanka
Atwani, Leon Gommans, Cees de Laat, Chrysa Papagianni, Ana Opresc [slides]
- SCinet Theater presentation: "Secure Collaborative Model Training
with vfl in multi-domain environments", Anestis Dalgkitsis Alexandros
Koufakis.[slides]
|
Demo:
Microservices-based FABRIC implementation of Collaborative Model Training Scenario with Vertical Federated Learning.
The European Union Aviation Safety Agency (EASA)
suggests that AI-driven algorithms, combined with fleet data, can detect
engine failures early, enabling proactive maintenance and enhancing
flight safety. The Independent Data Consortium for Aviation (IDCA)
supports data sharing to improve these algorithms. However, issues like
privacy, intellectual property, and regulations hinder this progress.
Overcoming these challenges could lead to safer, more reliable air
travel, benefiting society by reducing accidents and minimizing flight
disruptions.
With this paper we demonstrate how a Federated
Learning (FL)-based solution of collaborative model training could be
deployed in a multi-tenant and multi-domain network, using
microservices. The utilized DYNAMOS middle-ware, allows third-parties to
maintain control of their data, privacy concerns are taken into account
by design, and contractual agreements are enforced. We are proving our
solution in a transatlantic slice (USA-NL) deployed in FABRIC, to
simulate how airlines can exchange data and collaboratively train a
model detecting engine failures.
|
DYNAMOS and FABRIC
|
SCinet contributions, team members:
|
|
|
