The AI Research and Development Team at Google has made GPipe framework which is responsible for building large-scale and accurate deep neural networks open source. In short, GPipe is basically a scalable machine learning library.
This machine learning library is aimed to enable users to train large-scale deep neural networks faster, more accurately and potentially with less compute power. This move seems to be a part of a trend in which big vendors like Google, AWS and Facebook are seen pushing out open source AI development tools.
TEACHING A NEURAL NETWORK
The GPipe is said to split training into 'mini-batches' in order to determine the model error and then into even smaller 'micro-batches'. By using GPipe you can use a faster, more accurate and less memory intensive way for training deep neural networks which could be something positive for researchers.
OPEN SOURCE TREND
This is said to be one of the company's most important public contributions which is a software library for AI and machine learning.
This is something which enables the developers to create machine learning models which keep the user data anonymous. It uses techniques based on the concept of differential privacy. This already seems to be in use in many AI-based products and services.