Graph Representation Learning

Over at the day job, we've taken a keen interest into Graph Representation Learning and how these techniques can help us solve problems in financial services such as Fraud, Anti-Money Laundering and more. Here is some of the work we've been getting out lately to describe our approaches and challenges in applying these ideas at Capital One.

• This blog post on the Capital One site provides a good overview of the techniques we've explored and how we're applying them to finacial problems for the first time. This is a gentle introduction for a broad audience.
• Our paper at ICMLA 2019, Embedding Graphs of Financial Transactions, provides much more technical detail about our approaches.
• Doing this stuff on graphs with hundreds of millions of nodes and billions of edges can turn out to be tricky. Read our KDD MLG Workshop paper Graph Embeddings at Scale .
• Understanding and interpreting graph embeddings is another interesting challenge - it's hard to know what is "good enough" and how to understand the final representations. Our paper, On the Interpretability and Evaluation of Graph Representation Learning , at the NeurIPS Graph Representation Learning workshop puts forth some new ideas in this area.