Highly efficient and flexible Deep Learning building blocks for Arm and Power processors

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

The Prisma grant from Prof. Dr Alexander Breuer, CZS Endowed Professor of Scalable Data and Computationally Intensive Analysis, is part of a collaborative project to reduce the energy requirements of Deep Learning through the use of Tensor Processing Primitives (TPPs).

Goals

Deep Learning (DL) has established itself within the last decade as an irreplaceable tool for a multitude of problems. However, DL-based methods require an immense amount of computing power and energy. Innovations that accelerate the training and inference of deep neural networks are therefore in demand.

The goal of Professor Breuer's research project is the prototypical introduction of tensor processing primitives (TPPs) as building blocks for deep learning operations. Unlike other approaches, TPPs are portable and flexible and can be implemented on any hardware. The overall goal is to optimise execution times through the use of TPPs and thus reduce the energy requirements of Deep Learning.