- Keywords: deep reinforcement learning, resource allocation, actor-critic
- Abstract: Access to massive computation allows researchers and developers to succeeded in using technology to enhance processes in many applications. However there have been claims as to the tapering of the exponential decrease in the cost of hardware (following Moore's law) due to physical hardware limitations. Next generation special purpose systems making using of multiple kinds of coprocessors, known as heterogeneous system-on-chips, have been in active research recently. In this paper, we introduce a method to intelligently schedule a stream of tasks to available processing elements in such a system. We use deep reinforcement learning which allows for complex decision making and demonstrate that machine learning can be used for scheduling decisions and provides for a viable, likely better alternative to reducing execution time, given a set of tasks.