This thesis explores the integration of Bayesian neural network components into sensor fusion models to explicitly represent uncertainty under changing sensor configurations. By modeling fusion parameters as probability distributions rather than fixed weights, the approach enables the estimation of epistemic uncertainty when sensor inputs differ from those seen during training.
Aufgaben
- Identify and understand relevant literature on Bayesian neural networks and sensor fusion.
- Integrate a Bayesian neural network component into an existing sensor fusion model.
- Run a benchmark with different sensor configurations to assess whether configuration uncertainty can be captured by the Bayesian component
Anforderungen
Studiengänge:
- Computer Science
- Artificial Intelligence
- Robotics
- Math, Data Science or comparable degree program
Studienschwerpunkte:
- Software development / programming
- Machine learning / deep learning
- Digital image processing
- Statistics / data science
Fachkenntnisse:
- Solid understanding of popular machine learning and deep learning concepts
- Experience with statistical models
- Experience with 3D data processing and sensor data fusion is a bonus
IT-Kenntnisse:
- Confident use of Microsoft Office, Git, and Linux (Ubuntu)
- In-depth knowledge of Python, C, or C++
- Proven experience with popular machine learning frameworks
Sprachkenntnisse:
- English (fluent spoken and written)
- German is an advantage
Soft Skills:
- High level of initiative
- Strong analytical skills
- Structured and independent way of working
- Ability to work in a team
- Goal-oriented