mebeing
A new area of the machine learning research is deep learning. This had been introduced with the purpose to move machine learning closer to artificial intelligence. Although it takes days to train one of the systems to adopt deep learning but Google has found a way to fix this issue.
Recently, Google has announced to release Tensor2Tensor which is an open source tool to quickly train deep learning systems with the help of TensorFlow. By using just a single GPU, cutting-edge results can be attained in case of the best training model. This is a technique that can easily make an ordinary PC achieve results that were previously only possible with supercomputer-level machinery.
The system is very flexible as the standard and modular interface allows users to use virtually any training model, parameters, or data sets. The users are not required to change everything for changing just one component. This is an open source model which makes the community to share its own models for helping beginners.
Tensor2Tensor may not be that easily used at home as it requires good knowledge for deep learning. But this will definitely open ways for researchers to train their systems quickly that cannot be used with numerous GPU set-ups to training their deep learning systems.
Via: Engadget
