(Rice University) Computer scientists from Rice University and Amazon, using a divide-and-conquer approach that leverages the power of compressed sensing, have shown they can train the equivalent of a 100 billion-parameter distributed deep learning network on a single machine in less than 35 hours for product search and similar extreme classification problems.

Original source: https://www.eurekalert.org/pub_releases/2019-12/ru-rar120919.php