Safe and Robust Machine Learning - Application / Co-Design Perspective

Context

Machine learning is a promising approach for object recognition or segmentation in diverse domains. However, with the increasing use of machine learning models in safety-critical applications, it is also important to ensure safety/reliability as well as to increase the robustness of the ML models. From the hardware side, important aspects here include protection against transient random errors. From the software side, especially the one estimation of the uncertainty of the neural networks is interesting.

Aim of the work:
The goal of this work is to improve the robustness and reliability of machine learning models and to estimate the uncertainty of predictions. The focus is on the application and co-design perspective. Techniques for robustness and uncertainty estimation of neural networks will be developed and evaluated.

Tasks

• Review the current literature on robustness and uncertainty estimation of machine learning models.
• Design and implementation of techniques for robustness and uncertainty estimation of neural networks on hardware level
• Experimental evaluation of the developed techniques using benchmarks and real data sets
• Discussion of the results and derivation of conclusions

Prerequisites

• Strong motivation for the topic of machine learning and independent problem solving skills
• Basic knowledge in ML and neural networks is an advantage
• Programming experience should be available, Python knowledge an advantage but not mandatory.