Oct 30, 2019 · newsletter

Exciting Applications of Graph Neural Networks

Graph Neural Networks (GNNs) are neural networks that take graphs as inputs. These models operate on the relational information in data to produce insights not possible in other neural network architectures and algorithms.

While there is much excitement in the deep learning community around GNNs, in industry circles, this is sometimes less so. So, I’ll review a few exciting applications empowered by GNNs.

Overview of Graphs and GNNs

A graph (sometimes called a network) is a data structure that highlights the relationships between components in the data. It consists of nodes (or vertices) and edges (or links) that act as connections between the nodes. Such a data structure has an advantage when dealing with entities that have multiple relationships.

Graph data structures have been around for centuries and their modern use cases are wide. Well-known industry applications of graphs include the social networks of Facebook and LinkedIn, and street and road networks used in navigation apps such as Google Maps and Waze.

A graph data structure (image credit)

Graph Neural Networks are inspired by deep learning architectures, and strive to apply these to graph structures. Many of these architectures are direct analogues of familiar deep neural net counterparts. These include Graph Convolutional Networks, Graph Encoders and Decoders, Graph Attention Networks, and Graph LSTMs.

GNNs have been around for about 20 years, and interest in them has dramatically increased in the last 5 years. In this time, we’ve seen new architectures emerge, novel applications realized, and new platforms and libraries enter the scene.

Below, I highlight three novel uses of GNNs.

Application 1 - Predict Side-Effects due to Drug Interactions

Every year in the United States, hundreds of thousands of seniors are hospitalized due to the negative side effects of one or more prescribed drugs. Meanwhile, the number of older people prescribed multiple drugs at a time is expanding. Given the proliferation of pharmaceuticals, it is not possible to experimentally test each combination of drugs for interaction effects. In practice, doctors rely on training, understanding a patient’s medical history, and studying literature about the drugs in use to gauge the risk of harmful side effects.

Classification and similarity algorithms have been applied to this problem before, producing interaction scores. These results have been limited for a few reasons. The scores they produce are scalar values that highlight the risk of interaction without characterizing the nature of the interaction. These algorithms are limited to pairs of drugs.

By applying a type of GNN called a Graph Convolutional Network (GCN), a team at Stanford has been able to produce a model that can predict specific drug-drug interaction effects due to the interaction of more than 2 drugs. This model, which outperforms previous methods in identifying such effects, can identify side effects that are not attributed to the individual input drugs.

Example of input graph of drug and protein interactions, and side effect edges used to train the Stanford model (image credit)

Application 2 - Node Importance in Knowledge Graphs

The knowledge graphs produced by some enterprise companies are multifaceted, containing context and relationships across several types of entities and objects. Such graphs can contain billions of objects. Amazon is one such company, using knowledge and product graphs to capture the relationships between product data and the critical context that humans have but machines lack. Such graphs enable machines to excel at downstream applications like product recommendations and question answering.

However, at the scale of data that an enterprise uses, sifting through this massive amount of context can time-consuming and will impede graph-enabled applications.

To ameliorate this, Amazon has developed a GNN, called GENI (GNN for Estimating Node Importance), to distinguish the trivial facts and data from critical information contained in a knowledge graph. This algorithm was tested on knowledge graphs of movies, music, and general facts - but has wide ranging implications when applied to large scale graphs.

Application 3 - Enhancing Computer Vision with Physical Intuition

Computer Vision has advanced rapidly with the help of deep learning - in areas of image classification, object detection, and pixel segmentation (among others). Machines can distinguish and identify objects in images and video. There is still much development needed for machines to have the visual intuition of a human.

One type of human intuition is related to physics. If we see a ball bounce, we can reason about its trajectory. Even for more dynamic interactions among several objects, we can make reasonable predictions about what will happen without having a deep knowledge about the laws of motion or Newton’s three laws.

Interactive Networks are now giving machines this same physical intuition. This results in models that can predict what will happen over an extended time, given a few frames of a video scene. An example from DeepMind combines a CNN that distinguishes objects in a scene with an Interactive Network, which reasons about the relationships between these objects. The results of these models have been posted in videos, which are overlaid on CIFAR images. These videos compare the projected trajectories of objects against those predicted by a physics simulator.

Examples of trajectories done by a physics simulator (middle), compared with predicted trajectories over 40-60 video frames (right) (image credit)

Closing Thoughts

The applications discussed above highlight the functional variety of GNNs. In the first application, the GNN solution is used to predict graph edges. In the second, GNNs are used to score the importance of graph nodes. The third application uses GNNs predict the future states of a network.

These 3 applications only scratch the surface of the work being done around GNNs. The growth in this area is accelerating as people try to expand the usage of their existing graph-data. Researchers are also taking data that is in relational or document form, and reframing it in graph structures to take advantage of GNNs and other graph-analytical tools and methods. Though the above examples use GNN architectures for specific purposes, the potential to apply their solutions to similar problems in different domains is exciting.

Read more

Oct 30, 2019 · newsletter
Sep 27, 2019 · newsletter

Latest posts

Nov 15, 2020 · post

Representation Learning 101 for Software Engineers

by Victor Dibia · Figure 1: Overview of representation learning methods. TLDR; Good representations of data (e.g., text, images) are critical for solving many tasks (e.g., search or recommendations). Deep representation learning yields state of the art results when used to create these representations. In this article, we review methods for representation learning and walk through an example using pretrained models. Introduction Deep Neural Networks (DNNs) have become a particularly useful tool in building intelligent systems that simplify cognitive tasks for users. more
Jun 22, 2020 · post

How to Explain HuggingFace BERT for Question Answering NLP Models with TF 2.0

by Victor · Given a question and a passage, the task of Question Answering (QA) focuses on identifying the exact span within the passage that answers the question. Figure 1: In this sample, a BERTbase model gets the answer correct (Achaemenid Persia). Model gradients show that the token “subordinate ..” is impactful in the selection of an answer to the question “Macedonia was under the rule of which country?". This makes sense .. good for BERTbase. more
Jun 16, 2020 · notebook

Evaluating QA: Metrics, Predictions, and the Null Response →

by Melanie · A deep dive into computing QA predictions and when to tell BERT to zip it! In our last post, Building a QA System with BERT on Wikipedia, we used the HuggingFace framework to train BERT on the SQuAD2.0 dataset and built a simple QA system on top of the Wikipedia search engine. This time, we’ll look at how to assess the quality of a BERT-like model for Question Answering.
May 19, 2020 · notebook

Building a QA System with BERT on Wikipedia →

by Melanie · So you’ve decided to build a QA system. You want to start with something simple and general so you plan to make it open domain using Wikipedia as a corpus for answering questions. You want to use the best NLP that your compute resources allow (you’re lucky enough to have access to a GPU) so you’re going to focus on the big, flashy Transformer models that are all the rage these days.
Apr 28, 2020 · notebook

Intro to Automated Question Answering →

by Melanie · Welcome to the first edition of the Cloudera Fast Forward blog on Natural Language Processing for Question Answering! Throughout this series, we’ll build a Question Answering (QA) system with off-the-shelf algorithms and libraries and blog about our process and what we find along the way. We hope to wind up with a beginning-to-end documentary that provides:
Apr 1, 2020 · newsletter

Enterprise Grade ML

by Shioulin · At Cloudera Fast Forward, one of the mechanisms we use to tightly couple machine learning research with application is through application development projects for both internal and external clients. The problems we tackle in these projects are wide ranging and cut across various industries; the end goal is a production system that translates data into business impact. What is Enterprise Grade Machine Learning? Enterprise grade ML, a term mentioned in a paper put forth by Microsoft, refers to ML applications where there is a high level of scrutiny for data handling, model fairness, user privacy, and debuggability. more

Popular posts

Oct 30, 2019 · newsletter
Exciting Applications of Graph Neural Networks
Nov 14, 2018 · post
Federated learning: distributed machine learning with data locality and privacy
Apr 10, 2018 · post
PyTorch for Recommenders 101
Oct 4, 2017 · post
First Look: Using Three.js for 2D Data Visualization
Aug 22, 2016 · whitepaper
Under the Hood of the Variational Autoencoder (in Prose and Code)
Feb 24, 2016 · post
"Hello world" in Keras (or, Scikit-learn versus Keras)


In-depth guides to specific machine learning capabilities


Machine learning prototypes and interactive notebooks


A usable library for question answering on large datasets.

Explain BERT for Question Answering Models

Tensorflow 2.0 notebook to explain and visualize a HuggingFace BERT for Question Answering model.

NLP for Question Answering

Ongoing posts and code documenting the process of building a question answering model.

Interpretability Revisited: SHAP and LIME

Explore how to use LIME and SHAP for interpretability.


Cloudera Fast Forward is an applied machine learning reseach group.
Cloudera   Blog   Twitter