The project aims to provide a compiler extension and runtime for a new "paramserv" built-in function. In the context of large-scale machine learning, it will bring more performance when training a model with a data-parallel and model-parallel parameter server. SystemML already supports the data-parallel and task-parallel operator. Hence, we can focus on designing the parameter server primitive and implementing the parameter update strategies.

Student

Guobao LI

Mentors

  • Matthias Boehm
close

2018