A possible pathway to value for IoT

Internet of things (IoT) technology is coming into its own, enabled by improvements in network capacity and connectivity. However, to truly deliver widespread business value beyond alerting and monitoring, it needs to be intelligent. Unfortunately, the lightweight computers used in most IoT applications aren’t fit to efficiently perform the computation needed for intelligent capabilities. Modern deep learning models are characteristically large, requiring massive amounts of data and compute to deliver accurate results, but progress in binarizing these networks to make them smaller and simpler to compute without unacceptable sacrifices to network accuracy suggests real-world implementation in IoT devices is becoming feasible entering 2020.