Event Details

Abstract:

Money laundering is the deliberate concealment or disguise of the criminal origin of money or other assets. Frameworks such as AMLTRIX catalogue known tactics, techniques, and procedures, providing a shared foundation for detection and investigation. These public catalogues are also available to adversaries. Systems that depend on recognizing documented patterns therefore face a generalization problem as adversaries vary and recombine their methods: how can detection remain useful when the implementation of a concealment strategy changes?

This talk presents a proposed semi-supervised Bayesian framework that treats observability (the ability to reconstruct activity from available evidence) as an evolving latent state. Motivated by the Financial Action Task Force’s treatment of obscured financial activity as a potential red flag, the framework models how reconstructibility changes across successive transactions and transformations. It separately estimates whether lost visibility could be restored through accessible evidence and represents behavioral variation within those states. This provides a proposed basis for comparing patterns according to their effects on reconstruction and recovery as their implementations change.

The model couples stable, interpretable observability states with Bayesian nonparametric behavioral mixtures, allowing additional patterns to be represented while retaining a consistent vocabulary for reconstruction. A distinct recoverability variable differentiates equally opaque sequences according to their prospects for resolution. Known TTPs, domain knowledge, and partial annotations inform learning from largely unlabeled paths. The resulting posterior trajectories describe where reconstruction becomes uncertain, what evidence might resolve the gaps, and how a sequence relates to reference patterns. Loss of observability remains an investigative signal whose interpretation requires context.

The broader research question is whether reconstructibility and recoverability provide useful intermediate inference targets across domains. Potential applications include reconstructing cyber activity across endpoint, identity, and network logs; tracing consignments through fragmented shipping records; and reconciling orders, executions, and allocations across trading systems. Each presents distinct observation processes and evidence-access constraints. The talk presents the model’s architecture, an illustrative cryptocurrency sequence with a computed evidence update, and an evaluation strategy addressing investigative usefulness, generalization under adversarial adaptation, and the conditions required for transfer to other domains.

Speaker Bio

Casey King is Microsoft’s Chief Technology Officer for Strategic Partnerships in the Americas. He is also an Executive-in-Residence with the Center for the Future of Management at NYU Stern School of Business and an Adjunct Professor in its Department of Management and Organizations.

His research applies machine learning and Bayesian latent-state modeling to financial intelligence, adversarial behavior, and inference from incomplete evidence. He co-led a Defense Advanced Research Projects Agency seedling on financial warfare and has served as a principal investigator, co-principal investigator, and subject-matter expert on DARPA programs including XDATA, Quantitative Crisis Response, Anticipatory and Adaptive Anti-Money Laundering, and National Security Economic Theory.

For the U.S. Department of the Treasury’s Financial Crimes Enforcement Network, King built a platform that enabled jurisdiction-level analysis of approximately 100 million Bank Secrecy Act filings. His collaborative research with Mike Kane, undertaken for and reported by Barron’s, analyzed 24 billion U.S. equity trades, identified weaknesses in proposed stock-market circuit-breaker rules, and informed their subsequent redesign.

King spent a decade teaching at Yale University, including courses in big data, artificial intelligence, machine learning, anti-money laundering, and counter-terrorist financing at the Jackson School of Global Affairs and the Department of Statistics and Data Science. He was the founding Executive Director of the Yale Center for Analytical Sciences and has served as Senior Advisor to the Chief Data Officer of the U.S. Securities and Exchange Commission. Earlier in his career, he worked as a bond trader at Salomon Brothers. He holds a PhD from Yale University and a Master of Information and Data Science from the University of California, Berkeley.

ISI Speaker Host: Michael Collins, Senior Computer Scientist, Lead Scientist – Networking and Cybersecurity Division, USC Information Sciences Institute

Administrative POC: Katie Boeck, Office Manager – USC Information Sciences Institute / Networking and Cybersecurity Division