RocketML

From SEG Wiki
Jump to navigation Jump to search
ADVERTISEMENT

[1] RocketML is a HPC software infrastructure built from ground up to optimize every millisecond so that scientists and researchers can iterate on their machine learning experiments faster at lower total Cost.

In computer architecture, Amdahl's law is used to predict the theoretical speedup when using multiple processors. RocketML is built to scale by eliminating weak links. Every component is tuned so that the system is pushed to the limits of Amdahl's law. As a result RocketML outperforms all existing Distributed Machine Learning systems by orders of magnitude.

To build models with superior accuracy Data Scientists and Researchers need to iterate on their ideas without constraints. If a single iteration takes hours and days, training time becomes a deterrent to experimentation. Slow training is not an experience that is conducive to research and development of models. With increasing data sizes, flat computer performances, Distributed Machine Learning is an indispensable tool for researchers to get their job done. RocketML is THE most efficient distributed machine learning system in the market. It save researchers time & money with no limits on data sizes.

RocketML Distributed deep learning on CPU-based HPC clusters allows geophysicists to experiment with large 3D deep learning models without memory limitations as well as prohibitive wall-clock times for hyperparameter search. Training different network architectures with new datasets can be achieved in hours instead of days or weeks.