Gregor Schiele

The Elastic AI Ecosystem: An Open Platform for Embedded Deep Learning Research


Dr. Gregor Schiele is a professor for computer science at the University Duisburg-Essen (UDE) in Germany. He is leading the Embedded Systems Lab at UDE and is chairing the examination committee of the Master study program on Cyber Physical Systems. His research interest focuses on intelligent and adaptive systems, including embedded machine learning and self-organization for resource-restricted devices. He has published a multitude of papers and helped organizing many conferences and workshops over the years. 

Description of the Talk:

In this talk he presents the current state of the Elastic AI ecosystem, a holistic system for pervasive AI. It consists of (i) a hardware platform for exchangeable AI accelerators, (ii) an accelerator generation tool, and (iii) a distributed runtime system. AI-based applications can be composed out of different building blocks such as sensor data sources, preprocessing algorithms, and neural networks. Building blocks are executed locally (i.e., embedded), on nearby Edge or remote Cloud servers and compositions can be adapted at runtime, e.g., to switch from a Cloud AI to a local AI in case of networking problems. The talk introduces a current optimization approach used in the ecosystem to reduce the resource consumption of 1D convolutional networks.