11/2026 BIFOLD Colloquium

On the Different Flavors of Tabular Reasoning

Abstract:

The value of tabular data is clear: it fuels our understanding of, and decision-making in, many organizational processes. In this talk, Madelon Hulsebos will first address the need to assess the sensitivity of a (tabular) dataset, to avoid that it ends up in the wrong "hands". She'll revisit the notion of sensitive data, and argue that the sensitivity of data is defined by its context, and goes beyond personal identifiable information (PII). Building on this contextual notion, she will present a framework for assessing data sensitivity of any kind, which they developed in collaboration with the United Nations OCHA Humanitarian Data Centre. In the second part of the talk, Madelon will move from semantic to analytical reasoning over tabular data and introduce the SQaLe dataset, the first large-scale text-to-SQL dataset that is grounded in thousands of real-world database schemas. She will present its multi-purpose properties, and how it instruments our vision for small specialized text-to-SQL models as well as specialized models for multi-table retrieval in open-domain tabular insight extraction. The talk closes with a preview of their latest work around foundation models and agents for predictive reasoning over tabular data, and a reflection on the different characteristics and challenges that surface across these different flavors of tabular reasoning.