Abstract:
EDA is the iterative process of performing an action (such as a query on data), receiving the result, and deciding on the next step. The final goal of EDA is the extraction of insights, fragments of useful information justified by the data. Automating EDA requires overcoming several hurdles, notably identifying and correctly presenting the most interesting insights from a database. This talk will first highlight the many hurdles on the way to fully automated EDA as well as early works that attempt to solve this complex problem. I will then present a novel approach to automated EDA that draws inspiration from Operation Research. This approach enables massive scaling while retaining reasonable coverage of the search space.