Tuesday 12th of March, 14H00
Machine Learning for Critical Systems
Damien Fontanes, Dimitri Tombroff
Abstract
In real-life critical systems and applications, accessing the data to test, tune and run artificial intelligence algorithms involved several under-estimated problems.
The first challenge is mainly to provide consistent tools and execution engines usable for both data scientists and IT engineers. The inherent difficulty to act on large critical datasets is a second challenge, not to mention security issues.
Last of all, transferring the data to our research labs is most of the time forbidden considering Thales areas of expertise (Defense, Aerospatial, etc.), creating gaps between our available adaptative algorithms and production systems.
Several commercial products and technologies have started to tackle these issues, whereas working on production data remains a true challenge. In our talk, we will come back to a four years history of deploying large scale Thales cybersecurity solutions implemented on top of several open source technologies.
We will first explain the various strategies we put in place for deploying not just one but many such systems to deal with various kind of data.
Then we will present how we now jointly work in integrated teams mixing data scientists and data engineers to ease the access and work on large and normalized datasets.
Our primary focus will be to share real-life experience working on systems and applications, moving ahead from prototypes and MVPs.
