If the performance of automatic decision systems has now reached a level which attracts the interest of industries where safety is critical, the fact remains that they obviously cannot renounce the requirement to master automatic systems. In particular, learning algorithms should have a certain number of characteristics (such as specifiability, auditability, provability, etc.) which, even if not entirely sufficient, are necessary to increase our confidence in the system. Critical systems in the field of transport must undergo a demanding certification process before they can be implemented or used on the ground. The mandate of certification is to ensure that the program will operate in accordance with the intended environment. However, the certification of deep learning algorithms and computing 2.0 leads to new challenges.
This research them therefore aims to advance the state of the art with the objective of being able to certify systems, including components derived from learning algorithms, for use in the transportation industry.
Leader : Ettore Merlo, Polytechnique Montréal