Deep Learning with Dynamic Computation Graphs

TL;DR

This paper introduces dynamic batching, a technique enabling efficient batched training of neural networks with variable graph structures, and provides a high-level library for building such models.

Contribution

It presents a novel dynamic batching method and a library that simplifies creating and training graph-structured neural networks in static deep learning frameworks.

Findings

Enables batching over dissimilar graph structures.

Facilitates implementation of dynamic graph models in static frameworks.

Demonstrates concise, parallel implementations of various models.

Abstract

Neural networks that compute over graph structures are a natural fit for problems in a variety of domains, including natural language (parse trees) and cheminformatics (molecular graphs). However, since the computation graph has a different shape and size for every input, such networks do not directly support batched training or inference. They are also difficult to implement in popular deep learning libraries, which are based on static data-flow graphs. We introduce a technique called dynamic batching, which not only batches together operations between different input graphs of dissimilar shape, but also between different nodes within a single input graph. The technique allows us to create static graphs, using popular libraries, that emulate dynamic computation graphs of arbitrary shape and size. We further present a high-level library of compositional blocks that simplifies the creation of dynamic graph models. Using the library, we demonstrate concise and batch-wise parallel implementations for a variety of models from the literature.