Abstract: Large financial institutions across the globe have led the charge implementing artificial intelligence (AI) throughout their enterprises and are considered mature adopters of AI. However, many companies still struggle to rapidly expand AI use cases due to regulatory and internal data privacy restrictions for accessing customer data that contains PII. To address these concerns, MIT-IBM partnered with a major US Bank to generate a statistically similar, yet anonymous version of a credit card transaction dataset. The research partnership culminated in two neural network architectures for tabular time series: one for learning representations that is analogous to BERT and can be pre-trained end-to-end on downstream tasks, and one akin to GPT and is used to generate realistic tabular sequences, specifically credit card transaction data. Furthermore, this privacy preserved, synthetic dataset can be easily shared with data scientists for model development without jeopardizing private customer information. Throughout this presentation, we will discuss the technical architecture developed by MIT-IBM researchers, an accompanying visualization to better understand model output, and finally initial evaluation frameworks we're developing in conjunction with the US Bank to ensure applicability and establish industry benchmarks. Lastly, we will discuss the broader benefits of the MIT-IBM membership that provides industry with cutting edge methods and frameworks for building internal capabilities, while simultaneously providing researchers access to corporate data to help validate results and spark new research questions.
Bio: Austin Little is an AI Strategy Manager working on the Engagement Team for the MIT-IBM Watson AI Lab applying AI to the financial services and construction industries. Prior to working at the lab, Austin was a Strategy & Analytics Consultant at Deloitte working on AI strategy, proof-of-concept, and implementation projects across the healthcare, life sciences, and energy industries. Austin graduated with a degree in Physics from Morehouse College and conducted AI research at the Georgia Institute of Technology before focusing on applied AI for business and society