Accelerating Einsum Computations through Tensor Engines

Focus topic:	Artificial Intelligence
Type of funding:	Individual funding programmes
Funded institution:	Friedrich-Schiller-Universität Jena

The Prisma fund enables Prof. Dr. Alexander Breuer, Professor of Scalable Data and Compute Intensive Analysis at the University of Jena, to expand his expertise from the server to the mobile and embedded sector.

Goals

A large number of applications on mobile devices are limited by the available computing power and battery capacity. Smartphones and notebooks are examples of mobile devices that have been enhanced in recent years with special hardware components known as accelerators. Compared to classic components, accelerators have a high performance in some selected applications, for example in the evaluation of neural networks. There is therefore a strong need for highly optimized software that uses the available mobile accelerators with maximum efficiency.

The Prisma research project is investigating and optimizing the efficient evaluation of multidimensional linear algebra expressions on mobile devices. A particular focus is on novel and integrated accelerators that are tailored to artificial intelligence applications. The project is being carried out in close collaboration with industrial partners from the application area and the semiconductor industry such as Carl Zeiss AG.