Foundation Models for Structured Data

Workshop at the International Conference on Machine Learning (ICML) 2025

The first ICML workshop on Foundation Models for Structured Data (FMSD) will be held at the Vancouver Convention Center on July 18-19, 2025. We look forward to welcoming you in Vancouver!

Introduction

Structured data, such as tabular and time series data, is ubiquitous across countless high-impact real-world applications, from predictive analytics in finance and healthcare to climate modeling. Recent advances have led to the development of foundation models tailored to structured data. By learning from vast amounts of real and/or synthetic structured data, they exhibit strong generalization and the ability to adapt to new tasks with minimal fine-tuning.

Structured data foundation models are an emerging area of research undergoing rapid growth, yet they still remain critically under-explored relative to image and text modalities. So far, the different structured data sub-communities have had little opportunity to come together and share insights about how to build foundation models for structured data. Yet, strong synergies exist across modalities since models share similar pre-training and in-context learning paradigms. Furthermore, models trained on one modality can also demonstrate promising predictive performance in another.

The workshop on Foundation Models for Structured Data (FMSD) offers a place to jointly discuss foundation models for structured data, effectively addressing the gap and enabling the communities to capitalize on their synergies. We aim for advancements in foundation models that unify structured data modalities, addressing challenges of scalability and generalization across real-world applications.

Scope Clarification: We use the term structured data to specifically refer to tabular and time series data, and our focus is on predictive machine learning tasks such as tabular classification, regression and time series forecasting. For example, general-purpose graph-based methods are out of scope of this workshop, but foundation models for spatio-temporal forecasting that leverage graph-based architectures are considered in-scope, since the primary goal aligns with predictive structured data modeling. Another example is that tabular question answering system falls outside the scope as it does not focus on predictive tasks. To help guide submissions, here are a few clearly relevant prior works that align with the goals of this workshop:

  • Time series: Chronos, TimesFM, Moirai, Moment
  • Tabular: TabPFN, TabICL, TabForestPFN, CARTE

The key topics of this workshop include, but are not limited to:

  • Building Foundation Models for Structured Data
  • Datasets and Synthetic Data Generation Methods
  • Benchmarks of Structured Data Foundation Models
  • LLMs for Structured Data
  • Critiques on Structured Data Foundation Models
  • Real-World Applications of Foundation Models for Structured Data

Please see the Call for Papers for details.

Schedule

July 18th-19th 2025, Room TBD, Vancouver Convention Center

Refer to the ICML website for the detailed schedule.

Note: We do not yet know if the workshop will be held on July 18th or July 19th. We will update the date once we are given an update by the ICML organizers.

Detailed schedule TBD.

Contact

You can reach the organizers of the workshop at icml-structured-foundation-workshop@googlegroups.com.