Today’s edge computing technologies are designed to operate in very low-power environments with little connectivity. However, if AI algorithms, which typically require very high compute resources, can be designed and optimized to run at the edge in a low-power environment, this creates numerous possibilities for AI-powered IoT applications at the edge.
In this session, the possibilities for AI applications/use cases ranging from smart homes, smart factories to smart vehicles/driver assist, etc. are discussed. In addition, we look at real-world cases of AI/ML enabled through an AI co-processor/accelerator and an NN/Deep Learning algorithms.
Problems with AI at the edge can involve object detection from video, speech/voice recognition, or analyzing input from vibration sensors in machinery, to name a few examples. While some use cases may need compute infrastructure on cloud, others can be suitable for AI at the edge. These can be in surveillance/compliance or a host of other possibilities in smart home/factory/city scenarios.
Real-world use cases involving object detection using CNN on an AI engine in an FPGA processor is examined. And, challenges to implementing solutions to operate within the power/efficiency, latency constraints and FPGA footprint, with no significant loss in accuracy are detailed.