Congratulations to all participants on their ideas and especially their passion for science
The Federated Machine Learning Datathon, organised by TECNALIA, took place in October, with 16 researchers who had one thing in common – passion for data analysis technology and innovation, artificial intelligence and Big Data.
The competition involved a classic artificial intelligence problem under a federated strategy, and the implementation of a technological solution from a High Performance Architecture approach. Participants were given the rules and associated data when the competition started.
The main goal was to learn more about a current technology, “Federated Machine Learning”, competing between several teams of experts in Artificial Intelligence and High Performance Architectures; engineering, computing, maths and physics.
This federated learning enables clients who wish to protect their data to gather all their individual knowledge into a global metamodel, without compromising data privacy or confidentiality, i.e. avoiding sending raw data.
There are currently several promising line of research around this field, many of them linked to cybersecurity. AI research groups are starting to see the strengths and weaknesses of the “naive” approach of FML. The nature of the data belonging to each stakeholder can threaten the benefits of federated learning, which means that new breakthroughs by the scientific community are required.
In terms of architecture, FML is very attractive in this new scenario (edge/fog/cloud), where a good management of distributed resources is critical.
The datathon included two challenges: imbalanced data (not all categories to be predicted are equally represented in each stakeholder’s data), and a realistic technological solution to be implemented in the future.
