IAAILoopICL
LoopICL: Looping a single transformer block to solve tabular tasks
A tabular foundation model built from a single looped block with coupled cell and row streams refined by within-column and cross-column attention. Varying loop counts in pretraining and a learned exit gate let users trade inference cost for accuracy.
Inside the method
Table → [ Single block: cell stream ↔ row stream ] × r (exit gate) → predictions
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- Variable loops with a learned exit gate
- KV / state strategy
- Not applicable (in-context tabular model)
Reading note
Competitive with TabICLv2 on TabArena and TALENT at the same FLOPs while using nearly 90% fewer parameters.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{loopicl2026,
title = {LoopICL: Looping a single transformer block to solve tabular tasks},
author = {Amir Rezaei Balef and Katharina Eggensperger},
year = {2026},
eprint = {2609.36108},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.36108}
}