Abstract: Try to reason about something without any context. It’s possible, but your understanding will be limited and brittle. That’s because relationships between things give us critical information. In mathematics, we can model relational data as a graph or network structure -- nodes, edges, and the attributes associated with each. While deep learning has done remarkable things on Euclidean data (e.g. audio, images, video) graph deep learning has lagged because combinatorial complexity and nonlinearity issues making training very difficult and expensive. Yet it’s precisely the information hidden in that complexity that makes graphs so interesting. In this talk, Mark Weber will introduce a class of methods known as scalable graph convolutional networks (GCN) and share experimental results from a semi-supervised anomaly detection task in financial forensics and anti-money laundering. We will take a closer look at a new method developed at MIT-IBM called EvolveGCN, which uses recurrent neural network architectures (RNN) for handling temporal dynamism. We will discuss the implication of these results in anti-money laundering and beyond.
Bio: Mark Weber is an applied researcher and Strategy & Operations Lead at the MIT-IBM Watson AI Lab, a $250 million partnership funding over 200 scientists making fundamental breakthroughs in AI. Through the lab’s corporate membership program, which he runs, Mark works closely with global leaders across multiple sectors on the creative challenge of bridging fundamental science to real-world impact. Mark’s current applied research includes neuro-symbolic generative modeling for construction monitoring, graph deep learning for anti-money laundering, and supply chain demand forecasting. Mark also oversees strategic engagements with IDEO, the International Monetary Fund, and the Internal Revenue Service. Prior to IBM Research, Mark was a graduate researcher at the MIT Media Lab and a fellow at the MIT Legatum Center for Development & Entrepreneurship while he earned his M.B.A in finance from MIT Sloan. There he led the development of an open-source protocol called b_verify for verifiable records in supply chain finance. Before his foray into technology, Mark spent the first chapters of his career focused on political economy and development. He produced three documentary films on these subjects, most notably the critically acclaimed film Poverty, Inc
Applied Research Scientist | MIT-IBM Watson AI Lab