The Blue Horizon system comprises machines that autonomously enter agreements to execute workloads against microservices at the edge. Machine behavior is governed by user-defined policies and the Horizon peer-to-peer agreement protocol, and then logged to an Ethereum blockchain.
Simple Horizon workloads process realtime sensor data at the edge and publish derived insight to a central source for population analysis or machine retraining. Horizon's SDR Insight comprises three microservices, which tune a commodity Software Defined Radio (SDR) device and perform a number of functions. SDR-FM delivers radio station clips at the edge to IBM's Watson APIs for speech-to-text and sentiment classification. SDR-UI is a radio spectrum explorer application, with waterfall and audio clip capture capability. SDR-Aircraft tracks airplanes realtime, using radio transponder signals broadcast over ADS-B. Other workloads perform analysis of user internet connection quality (Netspeed) and air pollution particulate matter (Purple Air). NYU's Citygram application does realtime sound analysis on Horizon at the edge and contributes to NYU's study of noise pollution.
The Blue Horizon team is extending the platform to execute trained Machine Learning models on data at the edge. Both image recognition and audio data sampling workloads will be demonstrated at MWC. The platform enables data in both demonstrations to be processed at the edge for both security and performance.
Blue Horizon's Deep Learning box uses a number of Edge devices, acting as agreement agents (agbots), edge compute units, and data subscribers. In this demo, we use a pair of ODroid C2's as agbots to host a fully capable blockchain, a Raspberry Pi as a data subscriber, and an NVidia Jetson TX1 SBC as a deep-learning box for training and inference. The TX1 performs object classification at the edge, and the Pi runs audio classification microservices.
Blue Horizon Deep Learning Box: Jetson TX1, 2x ODroid C2, and 2x Raspberry Pi 3
(Laptop and Mobile Phone used as Visualization Devices)
The Deep Learning Box shown is a 3D-Printed base, with an acrylic cover available on Amazon.com. 3D Printing services provided by NYU's LaGuardia Studio in New York City. 3D Print files to be released shortly.
For more details on Deep Learning, Horizon hardware, the Horizon system, or the case itself, message us on our Discourse forum.
For more information on MWC2017 see mobileworldcongress.com