Blue Horizon Deep Learning comes to GTC 2017 in Silicon Valley
Blue Horizon, a Decentralized Autonomous Edge Compute system, presents cognitive workloads at the Edge using GPU-supported deep learning. The Blue Horizon project has emphasized support for Cognitive microservices at the Edge, supporting devices from Raspberry Pi's, Jetson TX1/2, x86, and cloud devices such as Power8 and Minsky.
On display at GTC 2017 was Blue Horizon's Deep Learning cube, with cognitive and traditional workloads:
- Audio Deep Learning on a Raspberry Pi 3 using a lightweight TensorFlow NN
- Inference at the Edge: Music, Speech classification
- Training in the Cloud
- Continuous training via active learning with at edge stream selection of useful training data
- Video Deep Learning on a Jetson TX2
- Motion identification using OpenCV
- Prioritization of images of interest: Histogram-based sorting / patch size / temporal priority
- Inference at the Edge: Caffe NN for image classification
- HA Microservices for lossless data transmission
- PMU / Synchrophasor data used as an example
- Supports US power grid throughput, latency, and high-availability needs
