The authors combined a frozen tabular foundation model (TabFM), which predicts on tables in a single forward pass, with an LLM-based agent that edits the surrounding data pipeline.
Language model agents evolve data pipelines to boost tabular foundation models
TabFM-Auto wraps frozen TabFM with an LLM agent that refines cleaning, features, context, and post-processing, achieving top Elo on TabArena and first on MLE-Bench-Tabular.
Big Tech
Deqing Fu · Huangyuan Su · Rajat Sen · Taman Narayan · Sujay Sanghavi · Abhimanyu Das · +1 more
Google Research · University of Southern California · Google DeepMind · Harvard University · University of Texas at Austin
Research Digest··2 min read
Fu et al.
Why this paper
From Google Research and 4 others
In one line
TabFM-Auto pairs a frozen tabular model with an LLM agent that iteratively refines the data pipeline, beating all competitors on 51 benchmarks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§