Zebra accelerates neural network inference using FPGA. User-defined neural networks are computed by Zebra just as they would be by a GPU or a CPU. Zebra is fully integrated with the traditional Deep Learning infrastructures, like Caffe, MXNet or TensorFlow.

There is zero FPGA knowledge required nor a single line of code to write to use Zebra. Simply link to the Zebra library to switch from CPU or GPU to FPGA in minutes. Zebra includes the FPGA image and the software stack, there is no FPGA compilation or FPGA tools to use.

This AMI contains the following components:
o Includes Zebra FPGA and software stack for Caffe and MXNet.
o This AMI supports your own CNN, similar to AlexNet, GoogleNet, Inception-v3 and ResNet-50 neural networks.
o Supports 1 FPGA (compatible with f1.2xlarge).
o Pre-installed versions of Caffe and MXNet.
o Ready-to-run graphical demonstration of image classification using CaffeNet and Caffe.
o Includes pre-trained ready-to-use neural network examples, including AlexNet, CaffeNet, GoogLeNet, Network in network, VGG-16, VGG-19, Inception-v3, Inception-v4, ResNet-50 and ResNet-152.

For more details :
o /home/ubuntu/Mipsology/V2017.10.1/doc/README.txt
o http://www.mipsology.com/aws/getting_started.html