Abstract: Fraud is an adversarial problem where attackers are constantly working against the system to identify and exploit any loopholes that might be present or suddenly appear. Unlike most machine learning applications where labels can be automatically inferred by the system, the underlying signatures of fraud attacks are often novel and varied and do not come with a label (even after the fact). This makes it critical for humans to work together with online decisioning to create an efficient human in the loop system to better discover blind spots and identify novel fraud attacks. In this talk, I will explain how Afffirm uses active learning to solve this problem that continuously ensures our systems are both effective in combating fraud and are built to scale with the growth of our business. Additionally, I will also discuss how Affirm uses unsupervised learning to improve fraud detection and quickly detect new fraud vectors.
Bio: Nitesh is the Head of Data Science at Affirm. In this current role, he is responsible for all the core modeling that runs the decisioning at Affirm, including identity, anti-fraud, credit, and personalization. Nitesh has over ten years of experience in analytics and machine learning with specific expertise in recommendation systems, pricing models, and targeted advertising. Nitesh obtained his PhD in mathematical finance where he applied modeling techniques to stock and options data. He is also passionate about explainable AI and data science for social good.
Nitesh Kumar, PhD
Head of Data Science | Affirm
devops-management-w19 | intermediate-w19 | machine-learning-w19 | talks-w19