System for ML

Recently, many researches use heavy deep learning models that need a lot of memory to train. The size of data items is also increasing such that it is difficult to increase the mini-batch size, which is used for the training of the models. This trend makes it difficult to train heavy deep learning models and large mini-batches on the commodity devices such as GPUs, which have extremely limited memory capacity compared to CPU memory capacity. 

The goal of system for machine learning  is to research new systems that can accelerate training algorithms or efficiently train heavy deep learning models on the small devices, distributed systems, and cloud computing systems.