Bayesian Machine Learning in practice

Bayesian Machine Learning enables the maximization of prior system information and provides insights into the steps the machine takes to reach its conclusions. These qualities make it useful in situations where data is scarce and where understanding the internal processes of the system is crucial for enhancing the solution's utility or justifying, for example, public policies. The tools and techniques of Bayesian ML are relatively recent and not so widely used, but the power of their results is already impressive. I will give a brief introduction to what Bayesian ML entails and show how its power grows when combined with professional knowledge of the system under study. I will detail how a system can be probabilistically modeled and, from it, unfolded to access its internal variables, where injecting prior information further enhances the results. I will provide two examples of work conducted with this tool. In one, we studied a mixture of populations within a community based on their collective electoral decisions, which allowed the client to optimize their corresponding strategies. In the other, we aimed to estimate early purchases in a food distribution system. In this latter case, we were approached after a multivariate proposal failed to achieve satisfactory metrics. I will demonstrate how Bayesian ML helped us understand why this happened, provided a solution, and offered the client a new and unexpected tool.



Note: This is in room 215.