Hardware Architectures for Edge AI using FPGAs
Project Principal Investigator(s): Shaik Mohammed Waseem, Prof. Subir Kumar Roy
Artificial Intelligence (AI) being part of this era’s rapid development in several forms like smart cities, smart agriculture, Industrial automation etc., has replaced the roles of human that usually involve an extent of difficulty and danger. But, to be realizable, AI depends on huge amounts of data to be provided as part of training and inference. Edge computing presents itself as an alternative to reduce the network infrastructure’s burden with respect to the data to some extent. Organizations like International Data Corporation (IDC) estimate that 55.7 billion Internet of Things (IoT) devices will be connected to the Internet by 2025 generating around 79.4 Zettabytes of data.
This project aims at the following:Ø Designing Field Programmable Gate Array (FPGA) based hardware architectures for training and inference of deep neural networks at the Edge.Ø The target domains are computer vision and autonomous control. As part of computer vision, Convolutional Neural Networks (CNNs) are considered for designing novel FPGA based hardware architectures and for autonomous control Reinforcement Learning (RL) based algorithms are considered. Ø Embedded FPGA boards like Avnet Ultra96 v2 (ZU3EGA484) and AMD Xilinx ZCU104 (XCZU7EV) that are based on Zynq Ultrascale+ MPSoC are used for carrying out this work.
