Matthias Jobst studied Electrical engineering (Diploma) in at TU Dresden with a main focus on digital chip design. Collecting first experiences still as a student at Rohde&Schwarz Messgerätebau GmbH and Coinbau GmbH, he started targeting the design of digital chips for highly efficient machine learning acceleration which led him to his current position pursuing a PhD in that area as a CeTI member at the Chair of Highly-Parallel VLSI-Systems and Neuromorphic Circuits at TU Dresden. There he was a main contributer in a team which won an award for energy-efficient AI hardware in the Germany-wide competition “Pilotinnovationswettbewerb ´Energieeffizientes KI-System´”. His current research focuses on making machine learning more energy efficient by utilizing and enforcing sparsity in neural networks and developing associated hardware accelerators.
Projects/Cooperation within CeTI you are involved in:
– design of a body computing hub, a chip to preprocess haptic data
– interfacing an ultra-high frequency wireless transceiver (with Chair for Ciruit Design and Network Theory)
With other rooms:
– Gesture recognition using Haptic Gloves for the „Hello Robot“ demo
– Gesture recognition for control of robots (with U2)
CeTI rooms within CeTI you are involved in:
What do you value most about your work at CeTI?
Working with researchers of so many varying fields of expertise gives the unique opportunity to learn about a multitude of topics.
What was your best moment at CeTI so far?
My first day of CeTI was at the General Assembly of CeTI. Getting to know and being welcomed warmly by so many people.
What else would you like to research?
Many machine learning applications are still not possible in small edge systems, such as wearables or smart devices – they just take too much energy and require huge memories for parameter storage. By compressing neural networks and making them only look at relevant parts of the data, I am trying to reduce these requirements.
How do you spend your spare time?
I go bouldering, sometimes also climbing and try to join some jam sessions with my piano.
ZEN: A flexible energy-efficient hardware classifier exploiting temporal sparsity in ECG data (Inproceedings)
In: Proceedings of the International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2022.
Event-based neural network for ECG classification with delta encoding and early stopping (Inproceedings)
In: Proceedings of the International Conference on Event-Based Control, Communication, and Signal Processing (EBCCSP), 2020.