Abstract

Nowadays, data science software and notably machine learning plays an increasingly important role in our daily lives. It thus becomes of paramount importance to have effective verification tools available to prove its correctness and reliability.

In this talk we will take you sightseeing through the architecture of Lyra, a static analyzer for Python components of a data science pipeline. The main attractions will be the submodules expressly designed to be stand-alone and reusable by other verification tools as well as the abstract domains designed to address new and challenging research problems that arise from targeting data science software.