SMART-DNN

Trusted AI embeddability is an important goal. It enables

1. Guarantee that the embedded model conforms to the development model
.
2. Estimate the performance (execution time, energy) of an AI component based on its parameters and the target hardware architecture
3. Identifying the impact of hardware faults on the behavior of ML components
4. Exploiting the regularity properties of ML algorithms on heterogeneous hardware architectures

The Smart-DNN project is part of the ConfianceAi project. It is being carried out in cooperation with the IRT systemX and the University of Lille.

The aim is to create a trustworthy environment to meet the challenges posed by the integration of artificial intelligence in critical systems.

This work is being carried out in cooperation with IRT SystemX, Safran and Thales. In particular, it aims to produce predictive AI models for estimating CNN model attributes as early as possible in the design chain.

This work is being carried out in cooperation with IRT SystemX, Safran and Thales.

Project funding includes Mr. Houssem Ouartateni's PhD thesis and a post-doc (currently being recruited), as well as a €70k support contract.

I am co-sponsor of the project with Professor El-ghazali Talbi (U. Lille). The project started in June 2022 and will end in December 2025.

I'm the co-leader of the project with Professor El-ghazali Talbi (U. Lille).