However, the reason that Tensorflow requires the machine learning model to be created in the form of a computation graph is such that the gradients can be backpropagated to enable the model to learn from the data. However, it will be interesting to see if these computation graphs in Tensorflow might also be used to parallelize learning to make it more efficient.
The ILP example reminds me of computation graphs in machine learning for Tensorflow. https://www.tensorflow.org/guide/graphs
However, the reason that Tensorflow requires the machine learning model to be created in the form of a computation graph is such that the gradients can be backpropagated to enable the model to learn from the data. However, it will be interesting to see if these computation graphs in Tensorflow might also be used to parallelize learning to make it more efficient.