===========================

The two cultures: Retrospects and prospects

Son P. Nguyen

University of Economics and Law, VNU-HCM

sonnp@uel.edu.vn

===========================

Abstract

In 2001, prominent statistician Leo Breiman wrote a renowned paper titled “Statistical modeling: the two cultures" in which he described two seemingly contrasting approaches to data analysis, namely

1. Data modeling: select generative models with emphasis on interpretability, and, to some extent, causality.

2. Algorithmic modeling: select models with best predictive capabilities (via validation) with little or no consideration to model explainability.

In this talk, after briefly reviewing Breiman's ideas, we will discuss some reasons why both approaches are needed in modern data science. Moreover, to get the best of both worlds, we would like a unified framework which allows data scientists to exibly take advantage of both types of modeling. Next, we will introduce fundamentals of probabilistic programming languages with a few examples to illustrate the combinations.

===========================

Keywords

Statistics, Machine learning, Data science, probabilistic programming language

#ISODS, #IUH, #DataiSG