At the Application and Test  Center AI, specific services are being developed as part of the Edge AI Performance and Testing (EdAPT) project that provide concrete information on the various implementation options at the interface between AI and hardware.

As part of this project, we are focusing on three key aspects:

Hardware performance and AI tasks

By compiling AI benchmarks from various partners and our own research, it is possible to identify the optimum hardware for processing AI tasks. In parallel to this, we have procured a wide range of hardware systems ­– including everything from small microcontrollers to powerful industrial PCs. These devices process and measure AI tasks. Key performance parameters such as processing time, latencies, resource requirements, and power consumption are then identified. Such insights allow the development of tailor-made advice and services for SMEs and other businesses.

Edge AI integration

As part of EdAPT, researchers are investigating the extent to which AI algorithms can already run on near-sensor hardware devices. Intelligent sensor systems are analyzed, tested in terms of their suitability, and trained for industrial applications.

Data reduction and edge computing

Data reduction is a key aspect when it comes to the use of edge devices. We are investigating various approaches to reducing data flow and evaluating their suitability for specific AI tasks. This work takes account of information loss and derives recommendations for handling raw data and intermediate information.

We want to enable companies to make sound decisions from the outset when it comes to identifying the hardware that best suits their AI applications.