Summary
Deep neural networks (DNNs) with billion- and trillion-scale parameters have demonstrated impressive performance in solving many tasks. Unfortunately, training a billion-scale DNN is out of the reach of many data scientists because it requires high-performance GPU servers that are too expensive to purchase and maintain.
This project will find ways at the compilers and operating systems level to make the training and use of large deep learning models accessible to mainstreamed data scientists by reducing their computational resource requirements.
