Showing 2 post.

Demo: Horizon Deep Learning, MWC 2017

Posted in news by Chris



Michael walks through edge audio classification with Deep Learning at Mobile World Congress 2017.  The demo shows Horizon and additional devices running edge Deep Learning workloads, within a single ecosystem.  Horizon, fully autonomously provides agreement management, and enables delivery of audio classification microservices running at the edge.  

Edge Devices: (2) Raspberry Pi 3's and an NVidia Jetson TX1 (video)

  1. Pi3 (Kato), with audio: USB sound card and analog mic
  2. Pi3 (Honeyman): Running two workloads: Horizon-Aural for audio classification, and Netspeed network analysis.  

Horizon-Aural (audio):

  • Edge audio processing using FFT's, prior to classification
  • Pi3 can run simple workloads on the Pi3, and much more on the Jetson TX1 or an x86 GPU-enabled machine. 
  • Audio classification with 4 classes: Music, Speech, Rain, Rock Music

Agbots: (2) ODroid C2's

  • Run Ethereum Blockchain, providing distributed ledger for device agreement and transactions
  • Serve as exchange for device identity registration, and services


Many thanks to Canonical for hosting us in their booth!


Horizon now supports Raspberry Pi 3

Posted in releases by Chris


Raspberry Pi 2 is still supported, and will continue to be supported into the future while we have participants and developers using Pi 2's.

Advantages of the Pi 3 over the Pi 2

  • Better WiFi connectivity:  Built-in WiFi (one more USB port for sensors / devices)
  • 1200MHz clock speed, up from 900MHz
  • CPU: Coretex A53, from Cortex-A7
  • GPU: 400MHz, up from 250MHz  (both VideoCore IV)
  • RAM: 1GB of 900MHz, up from 1GB 450MHz

source: writeup

All these features aid in Horizon's goal: Push smarter workloads to the edge, enabling the capabilities of modern computing to meld the edge and cloud.  Image processing, audio, and large data stream analytics. 

Magpi's Raspberry Pi 3 Specs and Benchmarks page