Unlocking the power of unsupervised learning with interpretable graph embeddings

Unsupervised learning has emerged as a powerful tool in machine learning, enabling the analysis of vast amounts of unlabelled data. At its core lies the creation of "embeddings," which are, in the context of graphs, essentially an efficient and AI-friendly way of representing graph properties. These embeddings can be utilised for various tasks, such as identifying patterns, making predictions, or grouping similar data points.

However, a key challenge in unsupervised learning persisted for our clients: the lack of interpretability. It is often difficult to understand how embeddings are generated and what information they capture, hindering their adoption, diminishing their trustworthiness, and complicating compliance with AI regulations. Therefore, addressing this challenge is critical for unlocking the full potential of unsupervised learning.

INGENIOUS: A Breakthrough in Interpretable Graph Embeddings

Euranova's research introduced INGENIOUS, a groundbreaking framework that addresses the interpretability challenge in unsupervised graph embeddings. INGENIOUS generates embeddings that are inherently interpretable, providing clear explanations for how they are derived. This eliminates the need for costly and time-consuming post-hoc analyses, which are typically required with other methods.

How INGENIOUS Works

INGENIOUS leverages data augmentation techniques to identify the most influential components of a graph, such as nodes and edges. By learning to distinguish between essential and non-essential elements, INGENIOUS provides insights into the decision-making process behind embedding generation. This transparency is crucial for building trust and aligning with the growing demand for explainable AI.

By enhancing performance, transparency, efficiency, and compliance, it empowers organisations to harness the full potential of graph-based machine learning: our clients gain state-of-the-art performance on a variety of downstream tasks, including graph and node classification, together with a clear understanding of how their embeddings are generated, promoting trust and confidence in the results. By eliminating the need for post-hoc analyses, they save time and resources. And finally, INGENIOUS facilitates compliance with AI regulations by providing interpretable and explainable results.

Applications Across Industries

This ability to understand and interpret graph embeddings unlocks powerful insights across a wide range of applications. It is particularly well-suited to address challenges in diverse fields where understanding the "why" behind the results is crucial. Here are some key examples:

• Fraud Detection: In financial transactions, it can analyse transaction graphs to identify suspicious patterns and anomalies, helping investigators to understand the underlying mechanisms of fraud and take proactive measures.
• Retail Analysis: it can analyse customer purchase history (represented as a graph) to understand product relationships and customer behaviour.
• Telecommunications: Churn prediction is a major challenge for telecom companies. By analysing call data records as a graph, INGENIOUS can help to identify patterns that precede customer churn, allowing companies to understand the reasons behind churn and take proactive steps to retain customers.
• Biochemistry and Drug Discovery: INGENIOUS can be applied to the complex graphs of gene expression pathways, protein interactions, and molecular structures. This can lead to a better understanding of disease mechanisms, identification of potential drug targets, and more efficient drug development.

This is just a glimpse of the potential applications of INGENIOUS. From identifying suspicious patterns and anomalies in financial transactions to understanding the dynamics and relationships within social networks, INGENIOUS has the potential to revolutionise various fields. Should your organisation need guidance to unlock the power of its graph data, do not hesitate to contact us.