Graph Learning & Artificial Intelligence

What is graph learning?

Simply said, it’s the application of machine learning techniques on graph-like data. Deep learning techniques (neural networks) can, in particular,  be applied and yield new opportunities which classic algorithms cannot deliver.

Graphs exhibit, like any other type of data, patterns which can be learned or detected. At least, provided there is enough data to learn from. Patterns can appear in different styles because a graph has intrinsically two types of elements:

  • links can appear (or not) in function of other links and nodes and one can predict whether or not a link should be present. Think of social relationships or references in a book, they do not show up randomly but according to the context.
  • nodes usually carry a payload (so-called property graphs) and the link structure in function of the payload can be learned and predicted. Think of job-titles and professional peers.

Graph learning (aka GraphML) can be applied to any type of graph but is especially useful when dealing with big-data. To name a few, large-scale analysis of customer and marketing data, combined with social network information reveals patterns which cannot be detected by tabular data alone. Forensic intelligence based on historical records combined with real-time information can predict critical events. If one integrates temporal and geospatial data in graphs one gets a very complete view of fraud patterns and transaction anomalies.

Applications of GraphML

Interaction networks can be trained to reason about the interactions of objects in a complex physical system. It can make predictions and inferences about various system properties in domains such as collision dynamics (rigid and non-rigid). It simulates these systems using object and relation centric reasonings using deep neural networks on graphs.

The nano-scale molecules have an inherent graph like structure with the ions or the atoms being the nodes and the bonds between them, edges. GraphML can be applied in both scenarios: learning about existing molecular structures as well as discovering new chemical structures. This has had a significant impact in computer aided drug design.

Zero-shot image recognition relies on the existence of a labelled training set of seen classes and the knowledge about how each unseen class is semantically related to the seen ones. One approach is to leverage structural information, in the form of graphs. Knowledge graphs can provide the necessary information to guide this task. 

Reading comprehension (and language understanding in general) is one of the complex reasoning tasks performed by humans; the answer to a question may not be located in any single part of the extract but may need global inferencing. Representing the passage in the form of a graph using a neural networks helps in better connecting the global evidences.

Combinatorial optimization problems over graphs are a set of NP-hard problems. Some of these can be solved by heuristic methods. In recent times, attempt is being made to solve them using deep neural networks. Consequently, GraphML techniques are also being leveraged to operate over these graph structured datasets.