Computational Graph
===================

In this chapter, we look at the following question: How does a machine
learning system efficiently execute such a program on hardware? We can
break this down into three sub-questions: How do we schedule and execute
the model described by a machine learning program? How do we improve the
model scheduling and execution efficiency? And can we implement
automatic gradient computation for updating the model? The key to
answering these questions is computational graph technology. To explain
this technology, this chapter explains the following key aspects:

-  Computational graph basics
-  Generation of static and dynamic computational graphs
-  Common execution methods of computational graphs

.. toctree::
   :maxdepth: 2

   background_and_functionality
   components_of_computational_graph
   generation_of_computational_graph
   schedule_of_computational_graph
   summary