Machine Learning System Aware Optimization

ABSTRACT New computer systems have emerged in response to the increasing size and complexity of modern data sets. In order to ensure optimum performance, software approaches have to be closely matched with the basic features of systems. This research demonstrates the impact of system-sensitive machine learning using an optimizer lens, a crucial design and solution factor in the majority of machine learning problems. The exactness and convergence rates are traditionally measured for the optimization method. In contrast, a number of system-related variables are crucial to modern computing systems' overall efficiency. Specifications such as data or parameters for the device and higher-level meanings, such as communication and computer interconnections may be included. We propose CoCoA, an overall learning method that closely reviews and incorporates device parameters into the process and theory. We have shown the impact of CoCoA on two conventional distributed systems, that being the traditional cluster environment and the increasingly (founded) machine learning environment. Our results show that we get orders of magnitude quick, by combining system parameters and optimization techniques, to solve current machine learning difficulties. These empirical findings support the assumption that device parameters give more knowledge about the scientific performance.