Support vector machines are one of many techniques within the realm of classification problems in machine learning – given labeled training data that contains multiple classes (e.g. pictures of cats and dogs with associated labels), we define an algorithm to determine which class a novel data point belongs to (in the cats and dogs example, we’d like our algorithm to be able to tell us whether a new picture contains a cat or a dog). I’ll talk about theoretical results related to support vector machines that we recently published in Foundations of Data Science, which use a classical theorem from topology to characterize when a linear classifier is optimal. I’ll also talk about some of the work I do at Amazon Web Services (AWS) and will highlight how a particular type of function called a convolution has played a particularly useful role. Finally, as someone who has transitioned from academia to a data science career in industry, I’ll discuss what background and skills make a strong foundation to compete for and thrive in data science roles.