AI & Graph Technology: How Graphs Accelerate Machine Learning

Last week in our five-part series on AI and graph technology, we examined knowledge graphs, which offer context for decision support.

This week we continue giving you a glimpse into how a graph technology platform like Neo4j enhances AI with context with a look at how graphs offer greater efficiency of processing, and how graph-accelerated machine learning uses graphs to optimize models and speed up processes.

Graph-Accelerated Machine Learning: Context for Efficiency

Current machine learning methods often rely on data stored in tables. Machine learning on such data is resource-intensive at best. More than half of enterprise CIOs surveyed state that iterative model training is one of their greatest challenges in taking AI projects from concept into production.

Graphs provide context for improved efficiency for machine learning algorithms because data is already connected in the graph model, enabling relationships of numerous degrees of separation to be traversed and analyzed quickly at scale. Thus the name “graph-accelerated machine learning.”

Humans naturally connect related information.

As an example, consider how people think when asked, “What does this picture of a dog remind you of?” A human wouldn’t need to run an intensive program, such as a nearest neighbor classifier, to compare that dog to all possible objects. We would quickly identify it with mammals – not humans or inanimate objects – and then classify it as a dog.

When data is stored as tables, it takes numerous iterations to connect it. For example, filtering processes are inefficient when they manifest relationships as table JOINs that bog down data pipelines. Data science practices such as collaborative filtering tend to require many JOINs as the result of multiple tables, indexes and lookup requirements.


Scale is another issue in machine learning efficiency. Machine learning algorithms may run a calculation against all the data.

To avoid this, analysts may create subsets of data manually. These approaches tend to slow iterations down because they are either computationally intensive or require human involvement. A simple graph query accelerates this process by returning a subgraph containing only the needed data.

Conclusion

Graphs provide context for improved efficiency because data is connected, enabling relationships of numerous degrees of separation to be traversed and analyzed quickly at scale.

Next week, in blog 4 of our five-part series on AI and graph technology, we will look at how connected features are used in many industries and have been particularly helpful for investigating financial crimes like fraud and money laundering.