Note: This video may require joining the NVIDIA Developer Program or login

GTC Silicon Valley-2019 ID:S9581:Deep Neural Network Pruning for Efficient Edge Computing in IoT

Mohammad Jahanshahi(Purdue University),RihTeng Wu(Purdue University)
Learn about the efficiency of edge analytics obtained from comprehensive experiments on detecting cracks and corrosion from images. We'll describe how robotic systems that can autonomously collect and analyze data provide an alternative to the time-consuming and labor-intensive manual inspection practices now used for civil infrastructures. On-board data analysis capabilities for efficient inspection path planning are essential for autonomous data collection. As a result, computationally expensive deep neural networks must be incorporated into inspection robots, which have limitations in their computing and memory resources. We'll explain how our solution achieves quick inference and low memory demands through transfer learning and network pruning and discuss how it enables efficient edge computing in the context of Internet of Things.

View the slides (pdf)