Beyond IID: How General Are Tabular Foundation Models, Really?
For researchers developing tabular foundation models, this benchmark reveals that current models are limited to simple IID tasks, highlighting the need for progress on more demanding real-world scenarios.
The paper identifies that tabular foundation models are evaluated only on standard IID benchmarks, missing challenging non-IID, large, and high-dimensional scenarios. To address this, they introduce BeyondArena, a unified benchmark with 142 datasets, and find that existing tabular foundation models excel only on tiny-to-medium IID data, while tree-based and deep learning models dominate on non-IID, large, and high-dimensional datasets.
Foundation models for predictive machine learning on tabular data have recently gained significant traction in academia and industry. Research communities across disciplines are increasingly evaluating tabular foundation models on diverse datasets and tasks. However, these task- and discipline-specific evaluations remain largely inaccessible to model researchers because benchmark software and evaluation protocols are fragmented. As a result, model researchers rely on standard benchmarks, which are mostly defined for tasks where tabular foundation models already excel. The most challenging scenarios are excluded, limiting meaningful progress in the field by focusing on marginal improvements on IID data rather than on broader, more demanding challenges. To overcome this, we introduce BeyondArena, the first unified holistic benchmark for tabular data that supports diverse task types (IID, temporal, grouped), across sample size and feature dimensionality scales, with diverse feature types (with text, with high cardinality) from a broad range of disciplines. To enable unified benchmarking beyond standard benchmarks, we introduce Data Foundry, a Python framework and metadata schema for curating tabular datasets for predictive machine learning. Our results across 11 models and 142 curated datasets show that existing tabular foundation models excel on tiny- to medium-sized IID data, while traditional tree-based and deep learning models still dominate on non-IID, large, and high-dimensional datasets. BeyondArena guides model research for the most demanding challenges in tabular data, enabling progress towards truly foundational tabular models.