Towards Understanding Layer Contributions in Tabular In-Context Learning Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Balef, Amir Rezaei, Koshil, Mykhailo, Eggensperger, Katharina
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917092574363648
author Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
author_facet Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
contents Despite the architectural similarities between tabular in-context learning (ICL) models and large language models (LLMs), little is known about how individual layers contribute to tabular prediction. In this paper, we investigate how the latent spaces evolve across layers in tabular ICL models, identify potential redundant layers, and compare these dynamics with those observed in LLMs. We analyze TabPFN and TabICL through the "layers as painters" perspective, finding that only subsets of layers share a common representational language, suggesting structural redundancy and offering opportunities for model compression and improved interpretability.
format Preprint
id arxiv_https___arxiv_org_abs_2511_15432
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Understanding Layer Contributions in Tabular In-Context Learning Models
Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
Machine Learning
Artificial Intelligence
Despite the architectural similarities between tabular in-context learning (ICL) models and large language models (LLMs), little is known about how individual layers contribute to tabular prediction. In this paper, we investigate how the latent spaces evolve across layers in tabular ICL models, identify potential redundant layers, and compare these dynamics with those observed in LLMs. We analyze TabPFN and TabICL through the "layers as painters" perspective, finding that only subsets of layers share a common representational language, suggesting structural redundancy and offering opportunities for model compression and improved interpretability.
title Towards Understanding Layer Contributions in Tabular In-Context Learning Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.15432