Time and place: 14:00 on Zoom. Register online!

Speaker: Fabrizio Ventola, TU Darmstadt

Title: Robust probabilistic circuits for efficient and reliable predictions

Abstract: Probabilistic circuits are prominent tractable probabilistic models which can provide exact answers to a wide range of probabilistic queries in a tractable way. Given their sparse nature and the structural constraints enabling exact inference, it is challenging to induce these models in high-dimensional real-world domains such as time series and raw images. In this talk, I show how we can leverage spectral modeling and the clear probabilistic semantics of probabilistic circuits to learn models able to provide efficient and reliable predictions in these challenging domains, and how to make these particular models more robust to distribution shift and out-of-distribution data.

Bio: Fabrizio Ventola is concluding his PhD at the AI & ML Lab of TU Darmstadt (Germany), advised by Kristian Kersting. His research focuses on deep tractable probabilistic models and how to enable them to efficiently provide insights on big data collections, such as performing accurate and reliable predictions in the presence of noise and missing data. He co-organized the last three editions of the workshop on Tractable Probabilistic Modeling (TPM), which were co-located with UAI, and a series of seminars mainly centred on AutoML and intelligent systems for data management.