Conference Description:

In the first part, we will explore one of the core concepts in unsupervised learning, dimensionality reduction. Dimensionality reduction serves two main purposes. First, it reduces the computational complexity of working with very large datasets. Second, it removes the non-relevant information in a dataset, surfacing the information that matters most. We will use dimensionality reduction algorithms to build an anomaly detection system; specifically, we will build a system to detect credit card fraud without using any labels. Anomaly detection systems are widely used in industry today to detect all types of rare events such as fraud (e.g., credit card, wire, cyber, insurance), crime (e.g., hacking, money laundering, drug, arms, and human trafficking), and adverse events (e.g., financial market meltdowns, cardiac events, and spikes in online traffic). In the second part, we will explore one of the core concepts in unsupervised learning, clustering. Clustering is able to segment entities (e.g., users) into distinct and homogenous groups such that members of a group are very similar to members within the group but distinctly different from members in other groups. This group segmentation is possible without requiring any labels whatsoever and instead relies on separating entities based on behavior. For example, via clustering, online shoppers could be grouped into budget-conscious shoppers, high-end shoppers, frequent shoppers, seasonal shoppers, technophiles, audiophiles, sneakerheads, back-to-school shoppers, young parents, senior citizens, and millennials. To perform clustering well, good feature engineering is required. In this course, we will explore loan applications, perform feature engineering, and segment users based on their potential creditworthiness. We will also explore how clustering allows efficient labeling, turning unlabeled problems into labeled ones, opening up the realm of semi-supervised learning.