White Paper: Machine Learning Methods for Enhanced Reliable Perception of Autonomous Systems

Machine learning methods for enhanced reliable perception

As part of the ADA Lovelace Center for Analytics, Data and Applications, the Fraunhofer Institute for Cognitive Systems IKS has developed a technical white paper on machine learning methods for reliable perception of autonomous systems. In the white paper, the researchers address the following aspects:

  • Problems and solutions
  • Uncertainty estimation
  • Out-of-distribution detection
  • Object detection
  • Application of machine learning methods

Uncertainty estimation in deep neural networks

In this technical white paper, new methods for quantifying uncertainty in deep neural networks related to perception tasks as well as monitoring system and out-of-distribution approaches will be reviewed, developed and evaluated. Insights and context with respect to different perception tasks, such as object detection, will also be provided and linked to the approaches for estimating uncertainty. The main focus will be on safety-critical applications where these methods provide crucial knowledge to systems for avoiding high-risk behavior and increasing overall safety.

Download white paper to continue reading.

You have further questions? Feel free to contact us!

Contact Press / Media

Adrian Schwaiger

Research Engineer

Hansastr. 32
80686 Munich

Phone +49 89 547088-358

Sascha Rudolph

Contact Press / Media

Sascha Rudolph

Head of Business Development

Hansastr. 32
80686 Munich

Phone +49 89 547088-381