Marc Neu's Successful Doctorate: Real-Time AI for Point Clouds on Hardware Accelerators

The ITIV team joins Marc Neu in celebrating the successful completion of his doctorate.

In his dissertation, titled “Template-Based Deployment for Real-Time Deep Learning on Point Clouds,” Marc Neu addressed a key challenge facing modern data-intensive systems: the efficient use of powerful AI methods under strict real-time constraints. The work was conducted in the research group led by Prof. Dr.-Ing. Jürgen Becker. Prof. Dr. Torben Ferber served as co-advisor.

Point cloud-based neural networks are particularly well-suited for applications involving irregular sensor geometries and very large sensor networks. Until now, however, there has been no practical methodology for deploying these models in real-time trigger systems for large-scale scientific experiments. With DeePloy, Marc Neu developed a new approach that maps graph-based point-cloud networks to dataflow accelerators in FPGAs and heterogeneous system-on-chip platforms while meeting hard real-time requirements.

The effectiveness of this approach was demonstrated in three real-world applications of the international Belle-II experiment. The results show that modern graph-based AI methods can be successfully deployed even under the extreme demands of large-scale scientific experiments.

The dissertation ties directly into Marc Neu’s primary research areas at ITIV. These include the development of dataflow-based data acquisition systems, the hardware acceleration of graph neural networks, and the data-driven design of high-performance trigger systems for scientific applications. His work makes an important contribution to the transfer of modern deep learning methods to practical real-time systems and opens up new possibilities for future research experiments.

The ITIV warmly congratulates Marc Neu on this successful completion and wishes him every success in his future academic and professional endeavors.