Brian Merrill 

Brian Merrill is a Managing Director in the New York Data Analytics practice of Deloitte Transactions and Business Analytics LLP with more than 15 years of experience in data management, forensic analysis, and advanced analytics services for various clients including forensic accounting and financial investigations of fraud, corruption, and bribery. Brian has led data analytics engagements for clients analyzing hundreds of millions of records to identify and extract key business insights to better understand trends, issues and anomalies within a business to respond to corporate as well as regulatory inquiries. Mr. Merrill is a Certified Fraud Examiner (CFE) and Certified Analytics Professional (CAP) who specializes in fraud detection for both reactive and proactive monitoring needs. With a focus on systems and relational database analysis, his technical experience includes the design, development, and implementation of software solutions with extensive experience in data mining, data analysis, and machine learning/artificial intelligence. Brian has demonstrated expertise working domestically and internationally on sensitive client service matters regarding internal investigations, forensic and corporate investigations, intellectual property litigation, trade analysis and FCPA matters across a number of industries, while managing the privacy, information security, analysis and execution performed by large multinational teams serving these clients. Brian has also written expert reports regarding the technical evaluation and analysis of data to support and respond to legal matters.

Abstract

Applications of GANs for Anomaly and Fraud Detection in Employee Expenses

In order to discover underlying patterns and processes of complex systems, typical process mining algorithms heavily rely on historic observed event logs and therefore fail to capture the impact of potential unobserved behaviors from the system. Generative Adversarial Networks (GANs) and the adversarial training process can be leveraged in producing system variants to assist with approximating the underlying system behaviors. GANs have shown promising results in detecting anomalies and rare-events in complex systems in recent studies. They can assist in surfacing hidden system behaviors and propose potential future anomalies which may enable stakeholders to take required precautions to avoid or mitigate them. We will describe the approach of using GANs and the comparison to traditional anomaly detection techniques to detect potential fraud.