TensorFlow Graph: Decomposition

Project Goal:

Given a computation graph, map the computations on the set of devices such that the power/energy consumption can be optimized.

The stages of the project are as below:

• Taking an example computation graph, deciding the cost model on the basis of power/energy consumption.
• Deciding the graph decomposition algorithm on the given set of heterogeneous devices such that the cost model optimizes the power/energy.
• Generalizing the cost model for any arbitrary tensor flow graph.
• Generalizing the graph decomposition algorithm for arbitrary set of heterogeneous devices.