Safety and Deep Learning Algorithms

  • Published:11.4.2022

The growing success of deep learning algorithms in solving non-linear and complex problems has recently attracted the attention of safety-critical applications. While state-of-the-art deep learning algorithms achieve high performance in various simulated and real-life cases, there is no guarantee for the reliability requirements that safety-critical applications typically demand. So are deep learning algorithms useful in a safety-critical application?  To answer this question, one must understand the potential of deep learning algorithms and their relation to safety.

This report primarily focuses on the underlying cause of faults in a visual deep learning algorithm to analyze the safety concerns and shortcomings of the existing mitigation methods.

By explaining the technical side of the problem, it is easier to understand what practical methods can be utilized to mitigate the safety concerns and how effective they might be in preventing faults. Relevant studies and mitigation methodologies are referenced throughout the report to reflect the ongoing research. The shortcoming of each proposed solution is discussed to emphasize the importance of further research on the topic.

The aim is to provide a practical guideline to educate the interested parties on the current state of technical problems, existing solutions, and potential research subjects to compensate for lacking areas. The main outcomes of this report are:

  • Providing a practical, complete, and categorical list of possible faults and their underlying cause for a visual deep learning algorithm.
  • Providing potential state-of-the-art mitigation methods and discussing their shortcomings.

Report authors

  • Tampere University DSII School
  • Saeed Bakhshi Germi

Please login to access related FIMAnet content.

If your organization is a member of FIMA and you don't have an user account, click here to request login credentials.