29.05.2024
| DSC Team |
Dr. Christopher Metz Successfully Defended His Dissertation
As the first doctoral candidate of the DSC, Dr. Christopher Metz successfully defended his dissertation titled “Towards Sustainable Artificial Intelligence Systems: Enhanced System Design with Machine Learning based Design Techniques”.
Efficient and innovative computations of machine learning (ML) algorithms are particularly important nowadays, especially when it comes to autonomous driving, the Internet of Things (IoT), and edge computing. Convolutional Neural Networks (CNNs), a key ML technique, require significant computational resources. To meet these demands, ML accelerators like General Purpose Graphic Processing Units (GPGPUs) are often used. However, GPGPUs cause high power consumption, making the selection of the optimal accelerator a complex, manual Design Space Exploration (DSE).
Christopher’s dissertation, written in the research group of the DSC spokesperson and co-founder Rolf Drechsler, presents innovative approaches to optimizing the DSE process. It develops techniques to precisely predict the power consumption and performance of CNNs during execution. This enables developers to estimate the power consumption and performance of GPGPUs in the early development stages without needing real devices, significantly reducing the number of necessary prototypes.
Additionally, developers must decide whether an application will be implemented on an IoT device or in the cloud, considering the available network, bandwidth, and latency. The dissertation introduces a system that assists developers in these decisions by considering the available network, bandwidth, and latency, allowing for informed decisions with regard to energy and performance requirements.
The entire DSC proudly celebrates the successful doctoral defense on 29 May 2025 and extends warm congratulations to Dr. Christopher Metz on this remarkable achievement!
Updated by: Jennifer Nüchter
« back
29.05.2024 | DSC Team
Dr. Christopher Metz Successfully Defended His Dissertation
As the first doctoral candidate of the DSC, Dr. Christopher Metz successfully defended his dissertation titled “Towards Sustainable Artificial Intelligence Systems: Enhanced System Design with Machine Learning based Design Techniques”.
Efficient and innovative computations of machine learning (ML) algorithms are particularly important nowadays, especially when it comes to autonomous driving, the Internet of Things (IoT), and edge computing. Convolutional Neural Networks (CNNs), a key ML technique, require significant computational resources. To meet these demands, ML accelerators like General Purpose Graphic Processing Units (GPGPUs) are often used. However, GPGPUs cause high power consumption, making the selection of the optimal accelerator a complex, manual Design Space Exploration (DSE).
Christopher’s dissertation, written in the research group of the DSC spokesperson and co-founder Rolf Drechsler, presents innovative approaches to optimizing the DSE process. It develops techniques to precisely predict the power consumption and performance of CNNs during execution. This enables developers to estimate the power consumption and performance of GPGPUs in the early development stages without needing real devices, significantly reducing the number of necessary prototypes.
Additionally, developers must decide whether an application will be implemented on an IoT device or in the cloud, considering the available network, bandwidth, and latency. The dissertation introduces a system that assists developers in these decisions by considering the available network, bandwidth, and latency, allowing for informed decisions with regard to energy and performance requirements.
The entire DSC proudly celebrates the successful doctoral defense on 29 May 2025 and extends warm congratulations to Dr. Christopher Metz on this remarkable achievement!
Author: Jennifer Nüchter
« back