Paolo Morettin's interests lie in the intersection of machine learning and logics. In particular, he has been working on probabilistic reasoning and learning over hybrid continuous/logical domains. He investigated the use of SMT technology for marginal inference, how to learn constrained hybrid models from unlabelled data and what classes of problems admit tractable inference. Recently, he contributed to characterizing the classes of parametric models that can be learned optimally via maximum-likelihood estimation.

Pedro Zuidberg Dos Martires has been studying weighted model integration from a knowledge representation perspective, which allowed him to formulate WMI as a generalization of algebraic model counting and standard knowledge compilation techniques. Together with Samuel Kolb (and collaborators) he has pushed the frontier of inference algorithms based on probabilistic circuits, introducing novel concepts such as lambda-SMT and variable orderings for SMT formulas. Furthermore, he has pioneered the use of Monte Carlo techniques to compute integrals in the weighted model integration setting. This enables the computation of high-dimensional integrals. Together with the other tutorial collaborators he has authored a software toolbox including various WMI solvers and the field's first survey. His ultimate goal is to leverage weighted model integration to power inference engines of probabilistic programming languages in the discrete-continuous domain -- an endeavor in full-steam development.

Samuel Kolb was awarded a PhD scholarship from the Research Foundation -- Flanders (FWO), a selective scholarship for fundamental research, to work on constraint learning, probabilistic inference methods and constrained predictions. He completed his PhD in December 2019 for which he received the rare summa cum laude distinction. Samuel Kolb has published more than 11 papers at top-tier conferences and journals (Machine Learning Journal, AAAI, IJCAI, CIKM, UAI). He has been actively involved with the ERC advanced grant SYNTH project on automated data science of his promoter Prof. Luc De Raedt.
In 2018 Samuel, along with Martin Mladenov, Scott Sanner, Vaishak Belle and Kristian Kersting, introduced the use of knowledge compilation techniques for WMI, i.e., hybrid constrained probabilistic reasoning. Together with Pedro Zuidberg Dos Martires and others, he has since co-authored multiple important contributions that established knowledge compilation as one of the main techniques for solving exact WMI problems. He led the development of a framework for hybrid constrained probabilistic inference (http://pywmi.org) that aims to make solvers and tools more accessible to researchers. Together with Paolo Morettin and others, he combined his expertise on constraint learning with probabilistic inference to learn probabilistic machine learning models with explicit constraints from unlabeled data.
Learning hybrid constrained distributions from data allows WMI to not only be used as an "assembly language" for higher order probabilistic inference models but also as a target for probabilistic machine learning approaches. In 2021 he was awarded a grant (spin-off mandate) from VLAIO to apply the fundamental techniques developed throughout his career towards commercial applications. Since then, he has focused on translating the research on probabilistic hybrid inference to solve real world decision problems that involve probabilistic machine learning models.

Andrea Passerini's research focuses on the combination of learning, reasoning and optimization. He pioneered the combination of machine learning and SMT in a number of early works on interactive optimization and structured-output prediction. He later introduced Learning Modulo Theory as a learning framework formalizing this combination. Together with Vaishak Belle and Guy Van den Broeck, Andrea Passerini coined the term Weighted Model Integration (WMI) to indicate the extension of Weighted Model Counting to hybrid domains, in a seminal paper that started the line of work on WMI. Later contributions to WMI include approximate or special-purpose methods, exact methods based on predicate-abstraction, WMI learning as well as a WMI toolbox and a survey.