TabMF: Learning Dataset Representations with Tabular Foundation Models

Key points

• Representations of supervised datasets support higher-level machine-learning tasks such as dataset retrieval and distribution-shift monitoring.1
• TabMF equips a predictive TFM with an architecture-aware dataset-level readout and adapts the model end-to-end using contrastive post-training on stochastic views of synthetic supervised datasets.1
• We instantiate TabMF on TabICLv2 and TabPFN-2.5 and evaluate the resulting representations across six tasks on unseen real datasets.1
• The two TabMF encoders obtain the strongest aggregate performance among the evaluated learned and classical dataset representations, with particularly strong results in dataset retrieval, compact benchmark construction, and distribution-shift severity assessment.1

Sources (1)

• [1]TabMF: Learning Dataset Representations with Tabular Foundation Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 04:28 PM

  We introduce TabMF, a framework that repurposes pretrained tabular foundation models (TFMs) as encoders of supervised datasets.

  Representations of supervised datasets support higher-level machine-learning tasks such as dataset retrieval and distribution-shift monitoring.

Extractive summary: sentences quoted from the sources.