Madelon Hulsebos

I’m a researcher at CWI (tenured in April ‘26) where I lead the Table Representation Learning (TRL) Lab. I’m also an ELLIS Scholar and faculty MT member of the ELLIS unit Amsterdam, and Member of the Scientific Advisory Board of Prior Labs. Previously, I did a postdoc at UC Berkeley and obtained my PhD from the University of Amsterdam for which I spent time at MIT and Sigma Computing. Before academia, I spent 2+ years in industry working on ML-driven automated data science tools. Find my full CV here.

My research is focused on table representation learning (TRL), basically anything “AI for tabular data”, since tables prevail in the data landscape, contain valuable insights, and fuel important decisions in everywhere from governments and industry to healthcare. My vision is to democratize insights from tabular data ✨, and my research is centered around that (see below). My research has been supported by an NWO AiNed grant, BIDS-Accenture Fellowship, CWI internal funding, and industry sponsors.

In 2021, I set the goal to establish tabular data as a key modality for AI akin to images and text. To this end, and ever since, I’ve happily been driving Tabular AI ♥ initiatives besides my own research: I founded the Table Representation Learning (TRL) workshop series at NeurIPS and ACL, organized various related efforts at Dagstuhl, EurIPS, etc, and established the TRL research theme at ELLIS Amsterdam. I’m generally reviewing for various research tracks/workshops at e.g. NeurIPS, ICLR, ACL, and SIGMOD.

Some research interests and contributions in Tabular AI in line with my vision for democratizing insights from structured data:

Tabular reasoning: TARGET table retrieval benchmark, LLMs for table QA, question ambiguity in tabular data analysis.
Table semantics: semantic column type detection (e.g. Sherlock) and contextual sensitivity detection .
Dataset search: vision and insights on dataset search systems, proactive task-based dataset search DataScout.
Table embeddings & retrieval: Observatory on relational properties in table embeddings, DCTR retriever for complex queries.
Corpora for tabular AI: GitTables: 1M+ real-world tables, SQaLe: 500K+ text-to-SQL triples, SchemaPile: 22K+ DB schemas.

Other topics I am working on are: agentic data science, tabular predictive insights (e.g., tabular foundation models), and beyond.