There’s a great deal of anxiety concerning AI among the general public. What is too often overlooked are the great possible benefits of AI to us in present day, excluding possibly the irresponsible creation of school reports in English classes. Alban Zammit is a machine learning engineer who has been crucial in the creation and implementation of exciting and benevolent advancements for PayPal which strengthens the security of the online purchases so many people make in present day. Fraudulent activity is often baked in as a business cost for many companies, particularly in the online world. Mr. Zammit and his team have set their sights on leveraging AI to combat these activities and even spot new fraudulent trends and tactics as soon as (sometimes before) they emerge. Achieving this requires more than the skill of the most talented coders, it necessitates vision and a tenacious commitment to reinvention. There’s an element of Alban’s work that has much more in common with improvisational jazz than one might ever suspect. In this genre, he is a master.
Machine learning is enabling an exponentially faster progression in the combatting of online fraudulent activity. RMR (Rapid Model Refreshes) is essential to this advancement but requires the building of a complex data and AI pipeline which can automatically pull and process the data needed for retraining. Alban is the developer and owner of such a pipeline for one of PayPal’s core ACH models (known as Add Fund, which allows you to instantly expand your PayPal balance with an ACH withdrawal on your bank account). As a major contributor to the inclusion of Incremental Learning (or Continual Learning) in PayPal’s Next Generation Risk Ecosystem, Mr. Zammit is enabling the company to outmaneuver fraudsters at an expedited pace. Alban stipulates the importance of Incremental Learning in the design of RMR stating, “If PayPal’s risk team spots a new fraud tactic in a specific type of transaction or customer segment, incremental learning lets us quickly assemble fresh data for that particular area. Then, instead of retraining from scratch, we start with the existing model and fine-tune it with this new data. This selective training approach means the model isn’t ‘forgetting’ its previous knowledge but is simply adapting incrementally to address the latest threats.”
PayPal has already experienced massive performance improvements and a measurable dollar impact achieved by the design of Alban Zammit and his team on this project. As we continue to see intelligent technology embed itself into our lives, there is the realization that it can be used by those meaning well or by those meaning ill. Having the brightest minds on the side of benevolent contributions to society-at-large makes for a better world for us all. Considering all outcomes from many different perspectives is something the world could use much more of, and this tactic typically leads to the best outcome. Alban himself agrees with this stating, “There were times when my engineering team and I had differing viewpoints on the technical approach but I found that open discussions, sharing insights, and reaching consensus were invaluable. These moments not only enhanced our final product but also strengthened my communication and negotiation skills. This knowledge was invaluable when I later built ARM’s incremental learning (IL) pipeline, which needed to perform quickly and with technical excellence to maintain reliability. Knowing that my work helped the business respond quickly to evolving fraud threats was highly fulfilling. Today, I’m grateful to be part of a field where I can directly contribute to safer transactions worldwide, merging my passion for mathematics and technology with meaningful, real-world impact.”
Sharon Howe is a creative person with diverse talents. She writes engaging articles for WonderWorldSpace.com, where she works as a content writer. Writing allows Sharon to inform and captivate readers. Additionally, Sharon pursues music as a hobby, which allows her to showcase her artistic abilities in another creative area.