Özer Cem KELAHMETKELAHMET
Publications

Publications

Papers on Data Language Models and tabular AI, with abstracts, links and BibTeX ready to paste.

Papers

Data Language Models: A New Foundation Model Class for Tabular Data

Özer Cem Kelahmet · arXiv preprint arXiv:2605.06290 · May 2026

Every major data modality has a foundation model that understands it natively, except tables. The paper introduces the Data Language Model (DLM), a foundation model that reads tables the way a language model reads sentences: directly from raw cell values, with no serialization and no preprocessing. The first model of the class, with 140M parameters trained on more than 2.3 million synthetic and real tabular datasets, outperformed gradient-boosted ensembles, AutoML stacks and tabular foundation models on established row-level prediction benchmarks, and could name the industry sector of an unseen dataset from raw cell values alone.

arXiv abstract · PDF · Plain-language explainer

@article{kelahmet2026dlm,
  title     = {Data Language Models: A New Foundation Model Class for Tabular Data},
  author    = {Kelahmet, {\"O}zer Cem},
  year      = {2026},
  journal   = {arXiv preprint arXiv:2605.06290},
  eprint    = {2605.06290},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2605.06290}
}

An In-Context Imputation Head for Tabular Data: Chained, Validation-Gated, and Evaluated on Three Imputation Boards

Özer Cem Kelahmet · Paper draft · 4 November 2026

An 18.97M-parameter head on a frozen pre-trained tabular encoder fills missing cells column by column, in a chain, and a validation gate can always fall back to a MICE-equivalent chain or to the column mean. It leads MICE on a synthetic no-prior split and on 20 real datasets, is ahead of a published LLM-imputation study on its nine-cell mean but behind on MCAR, and catalogues seven harness and pipeline defects that each produced a convincing false result.

PDF · Plain-language explainer

@misc{kelahmet2026icihead,
  title     = {An In-Context Imputation Head for Tabular Data: Chained, Validation-Gated, and Evaluated on Three Imputation Boards},
  author    = {Kelahmet, {\"O}zer Cem},
  year      = {2026},
  note      = {Paper draft},
  url       = {https://www.ozercemkelahmet.com/papers/in-context-imputation-head.pdf}
}

Articles

Every article on this site has a Cite this article box under its title with APA and BibTeX. The BibTeX for all of them is in one file: articles.bib. All papers above are in papers.bib.

How to cite this site

Cite the paper for the Data Language Model idea and an article for a specific explanation or measurement. Articles carry their own publication date and a version history at the end, so a citation can name the version that was read.

Last updated Nov 18, 2026
Type to search.
    ↑↓ navigate↵ open