Dependable Deep Learning for Safety-Critical Airborne Embedded Systems

Important results from the project

The aim of the project has been to study machine learning-based safety-critical systems from different perspectives, which has been reported in scientific articles. For example, methods for neural networks in dependable systems have been investigated. Research on methods for correct selection of data and synthetic generation of data for training neural networks has been conducted. The results of the project are largely consistent with the goals set out in the project application.

Expected long term effects

A state-of-the-art conference paper in the field of machine-learning in safety-critical systems has been written. In addition, we have published two journal articles and 8 international conference papers. Saab has demonstrated the execution of deep learning on heterogeneous hardware for embedded systems, a work that is being taken forward in a new related project (FASTER AI). The project has produced 8 master theses. Expected effects of the project are deepened collaboration between the parties and strengthening of research in the field both at the company and the university.

Approach and implementation

In the beginning, the project was carried out according to plan. Once the project had been going on for a while and research areas had been identified, new opportunities for research were discovered and new applications were written. Through SSF, the main project manager was given the opportunity to do research in 80% of his time. MDU´s research profile, Dependable Platforms for Autonomous Systems and Control, meant that the research in SafeDeep could be made more visible. Personnel changes within the project led to a change in the research focus and the exploration of new research areas.

External links

List of publications related to the project SafeDeep projektsida, MDU