This seminar delves into the innovative approach of layer-fused processing for Deep Neural Networks (DNNs) on multi-core accelerator systems, presenting the Stream framework. We explore the design space to optimize the execution of DNNs by fusing consecutive layers, aiming for enhanced efficiency and performance. The discussion covers key aspects of the Stream framework, including its architecture, design considerations, and the impact on multi-core accelerator systems.
1/12/2023 11:00 - 12:00
ESAT Aula R