Embedded Deep Learning Choices

When it comes to running deep learning of ML applications we can find only a few which can support the matrix crunching efficiently. The Nvidia Jetson series especially "NVIDIA Jetson TK1 developer kit" is hugely popular. They employ the similar computing architecture which runs in the GPU stacks which does the deeply deep learning tasks for the Search giants like Google. Jetson is being discontinued and new powerful option is the product line being "NVIDIA Jetson AGX Xavier". Amazon Alexa also has development boards, but it sends data to the cloud and analytics is done in the cloud,so it doesn't qualify as a complete solution for deep learning for the edge. The next promising architecture is from ARM , the Mali GPUs.In that series they do have a development board "HiKey 970" which seems to be a complete solution for the Edge processing. Intel after acquiring Movidius has also gathered a powerful edge. Movidius Myriad2 processors is meant for vision and AI related prcessing. There is Development board called Neural Compute Stick which is with Movidius processors which support a lot of Machine Vision algorithms. There are interesting debates on GPU vs CPU and Analytics in the Cloud Vs Analytics in the Edge .Currently the growth rate of Machine Learning algorithms are much more than that of the supporting embedded platforms. Doing analytics on the cloud seems to be the right things to do especially when you are having a number of sensors connected to cloud(So called IoT !!). But when your application is very much specialized like a handheld device looking to find trace of a bacteria at a remote site , its better to Edge Compute.