Year: 2022
Type: dissertation
Abstract: We implemented a high-performance deep-learning-based video content recognition application.Meanwhile, we proposed a resource balancing algorithm to allocate proper computation resources to different DNN layers and achieved minimized latency.2. We developed a fine-grained inter-layer pipeline architecture to enable the overlapping of execution of DNN layer operations.3. We proposed a high-efficiency Winograd systolic architecture for convolution layers and explored the memory banking optimization in the Winograd systolic architecture.4. We proposed a novel algorithm that maximizes the computation unit efficiency of the hardware platform in DNN-related applications.
Author: Liu, Xinheng
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220795073