Madelon Hulsebos

AI for Good Webinar Series 'From Molecules to Models' - today with Madelon Hulsebos on 'What AI is leaving on the table: Towards democratizing insights from tabular data'.

💬 Webinar Series 'From Molecules to Models' - today with @madelonhulsebos.bsky.social 🇳🇱 on 'What AI is leaving on the table: Towards democratizing insights from tabular data'.

📅 Mon, Feb 23
🕓 16:00-17:00 CET

🔗 Registration: https://bit.ly/4ash9gn
ℹ️ More info: https://bit.ly/47Qclim

People can join also online! Check the page for the link.

Join the TRL Seminar tomorrow for talks on query ambiguity in open-domain data analysis @daniel-gomm.bsky.social, the SQaLE dataset for training text-to-SQL models @cowolff.bsky.social, and pattern/rule inference from tabular data @erkankarabulut.bsky.social.

Info: trl-lab.github.io/trl-seminar/

Today I wrote an elaborate email, which I hadn't done for a while. I realize that I start to enjoy crafting nice and authentic stories, word by word, now that most texts are LLM'd.

The LLM era makes me more creative and thoughtful about my daily writing than before.

Here's to more writing in 2026!

- Besides the workshop, I’m really looking forward to other great talks from the tabular community during the main program eg on TabArena, ConTexTab, the keynotes/panels as well as other great work presented there. Generally it’ll be amazing to connect with the European AI crowd!

- The workshop ends with a reception for the vibrant tabular AI community to connect over drinks: can’t wait to catch up with folks! Really grateful to our sponsors for accommodating this: SAP, Prior Labs, and ELLIS unit Amsterdam 🙏.

- Our contributions: 1) a take on ambiguity in NL-based tabular data analysis, spotlight+poster: arxiv.org/abs/2511.04584, 2) SQaLe: the largest text-to-SQL dataset grounded in real schemas, poster: openreview.net/forum?id=6Ps..., 3) my surprise keynote on our work & reflection on the community :)

- We’re organizing the AI for Tabular Data workshop w/ 3 keynotes, 9 spotlights, and 48 fresh papers in 2 poster sessions ✨ Excited about the great directions and quality of the work in this community! Papers & program at sites.google.com/view/eurips2.... Join us Sat 6 Dec at the Univ of Copenhagen!

I’m at EurIPS in Copenhagen this week! Great seeing the AI crowd come together in Europe 🤩.

Would love to chat with people about 1) AI for tabular data, 2) new collaborations (prospective students, phd visitors, peers from other labs/disciplines), 3) the future of AI in Europe.

Some highlights:

Looking forward to hosting Irina Saparina (Univ. of Edinburgh, saparina.github.io) this Friday for a talk in the TRL Seminar on "From Implicit Bias to Explicit Choice: Handling Ambiguity in Text-to-SQL Parsing" 👀.

📆 Fri 28 Nov 4-5pm CEST
📍 CWI lobby 3:50pm or Zoom 👇

trl-lab.github.io/trl-seminar

Paper: openreview.net/pdf?id=6PsKD...
Dataset on HuggingFace: huggingface.co/datasets/trl...
Blogpost on HuggingFace: huggingface.co/blog/cwolff/...

Really great effort that is led by @cowolff.bsky.social in colab with @daniel-gomm.bsky.social, and will be presented at the AI for Tabular Data workshop at EurIPS on 6 Dec! Stay tuned for our research on insight extraction from tabular data (also beyond text-to-SQL).

Excited to introducd the SQaLe dataset for text-to-SQL training/tuning at scale. SQaLe is the first large-scale text-to-SQL dataset grounded in real schemas and covers a broad range of realistic queries (eg # joins, operations).

SQaLe provides 517K high-quality question/schema/query triples 🤩.

We consider NL tabular query ambiguity through the lens of cooperative interaction. Read more about the insights and implications in the paper led by Daniel, which will be presented at the EurIPS workshop on AI for Tabular Data! arxiv.org/pdf/2511.04584

Really enjoyed this reflection exercise 🙂

For a long time, we reviewed and debated properties of queries in tabular QA datasets and practice, particularly in an “open-domain” context. Expecting “platonic queries”, map to necessary and sufficient data items and only the exact operational procedure to address the query intent, is unrealistic.

Are We Asking the Right Questions?

With @daniel-gomm.bsky.social and @cowolff.bsky.social we wrote a reflection on ambiguity in natural language queries for tabular data analysis: What queries should we expect, facilitate? On what should tabular data analysis systems be evaluated, and how? See thrd

I'm hiring PhDs through the ELLIS PhD program for research on table representation learning to democratize insights from tabular data!

Info: trl-lab.github.io/open-positio...
Apply: apply.ellis.eu (31 Oct)

Join us in beautiful Amsterdam at @cwi-amsterdam.bsky.social & @ellisamsterdam.bsky.social ✨

This exciting workshop is organized by @effylix.bsky.social, @lennartpurucker.bsky.social, Peter Baile Chen, Frank Hutter and me.

With talks by Marine Le Morvan (Inria), Floris Geerts (University of Antwerp) and @akhtarmubashara.bsky.social (ETH Zurich), and more TBA.

We hope to meet you there!!

Announcement for the EurIPS'25 Workshop on AI for Tabular Data. More information about the workshop can be found at: https://sites.google.com/view/eurips25-ai-td/home.

🚀 Excited to announce the AI for Tabular Data workshop at EurIPS 2025 in Copenhagen!

CfP: sites.google.com/view/eurips2... (papers due 20 Oct)

Join us @euripsconf.bsky.social to discuss neural tabular models and systems for predictive ML, tabular reasoning and retrieval, table synthesis and more ✨

Also hope attendees will enjoy the tutorial program, which I helped compile this year, with exciting tutorials on topics such as vector search, data discovery, graph databases, AI and relational data! Full list at vldb.org/2025/?papers...

Off to London 🇬🇧 for @vldb.bsky.social!

Looking forward to the panel discussion on neural models and tabular data Tue 10:45 🔥. I’ll share my thoughts on the state of affairs, promising paradigms, and future directions for TRL.

Will share my input and some take-aways of the discussions post conf.

We show that metrics like sacrebleu and bertscore aren't fit for tabular QA eval of LLMs as scores are inseparable. We also find a large gap between multiple-choice eval as in TQA-Bench and an LLM-judge which aligns with human annotation.

Bottomline: LLMs aren't robust for real-world multi-table QA

In our contribution led by @cowolff.bsky.social from the TRL Lab (trl-lab.github.io) we wondered: how well do LLMs reason over tabular data, really? We find that LLMs don't ack nor handle tables with disturbances (eg missing vals or duplicates) necessitating explicit prompting or cleaning pipelines.

How well do LLMs reason over tabular data, really? Large Language Models (LLMs) excel in natural language tasks, but less is known about their reasoning capabilities over tabular data. Prior analyses devise evaluation strategies that poorly reflect an...

The full program of the workshop is at: table-representation-learning.github.io/ACL2025/. Besides excellent contributed work (proceedings: aclanthology.org/2025.trl-1.0...), we'll have invited talks by Dan Roth, Tao Yu, Edward Choi and Julian Eisenschlos!

Excited to be at ACL! Join us at the Table Representation Learning workshop tomorrow in room 2.15 to talk about tables and AI.

We also present a paper showing the sensitivity of LLMs in tabular reasoning to e.g. missing vals and duplicates, by @cowolff.bsky.social at 16:50: arxiv.org/abs/2505.07453

The paper's called:
"How well do LLMs reason over tabular data, really?" 📊

We dig into two important questions:
1️⃣ Are general-purpose LLMs robust with real-world tables?
2️⃣ How should we actually evaluate them? (2/4)

Very excited about this! Aligns well with my 🇪🇺 focus this year.

Join us for discussions and talks on data management aspects for end-to-end ML on 27 June at @deem-workshop.bsky.social in Berlin. Keynotes by @pinartozun.bsky.social and @gaelvaroquaux.bsky.social 🤩

Check the full schedule deem-workshop.github.io#schedule & proceedings dl.acm.org/doi/proceedi...

Very interesting! I’ve seen synthetic data, with certain patterns (eg TabPFN with SCMs), come real far, but not entirely random compositions of tokens. Do you understand or have an hypothesis for what leads to this observation?

🚨What is SOTA on tabular data, really? We are excited to announce 𝗧𝗮𝗯𝗔𝗿𝗲𝗻𝗮, a living benchmark for machine learning on IID tabular data with:

📊 an online leaderboard (submit!)
📑 carefully curated datasets
📈 strong tree-based, deep learning, and foundation models

🧵