Deep learning has exploded in recent years. AI that can learn to create its own musical scores has been an area of particular interest. Training machine learning networks to have a basic understanding of musical composition by using existing music examples has been attempted by several researchers. Although many interesting results have been produced, the training times are extremely long and require powerful servers.
The advent of FPGA's that can be re-tasked to work with common AI frameworks such as Keras, Theano, Tensorflow, etc. allows training times to be significantly reduced. Adding ARM system-on-chips to FPGA silicon results and a self-contained system that can run a Linux operating system and AI framework on the ARM CPUs while leveraging the accelerated training FPGA engine on the same die.
My goal is to build a Linux OS that boots, connects to nearby units over a Wi-Fi mesh, and trains a music learning deepnet accelerated by the on-die FPGA.
Comments