Machine learning researcher advancing tabular foundation models (TabPFN): new architectures and FP8-efficient inference
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
No output score yet: the snapshot lacks a complete publication-volume record. This axis would measure total works, per-year productivity, and venue spread; only 4 works are visible in the recent three-year window.
Impact—
No impact score yet: citation-based influence cannot be scored reliably. h-index is 1 and i10-index is 0 in this snapshot — an early record too sparse for a stable measure of citation uptake and downstream use.
Focus0/100
Focus scores 0 in this snapshot. The field distribution splits evenly — 50% tabular deep learning and foundation models, 50% efficient inference in tabular deep learning — and OpenAlex topics spread across five clusters, led by Machine Learning and Data Classification (3), with database systems (2), neural networks (2), and two singletons. The catalogued work is thematically tight, but the field metric reads as distributed rather than concentrated.
Collaboration78/100
Collaboration scores 78, grounded in 49 distinct coauthors. The recurring cluster is a dense tabular-ML group: Frank Hutter, Noah Hollmann, Jonas M. Kübler, Klemens Flöge, Léo Grinsztajn, Nick Erickson, Oscar Key, Felix Birkel, Kürşat Kaya, and Adèle Frankel — consistent with team-based foundation-model research rather than solo work.
Activity47/100
Activity scores 47. The counts-by-year series shows all visible output dated 2026 (4 works), and recent3yWorks is 4 — a concentrated burst rather than a multi-year track. Both catalogued papers are 2026 arXiv postings, one first-authored, consistent with a recently started or rapidly ramping publication record.
Reproducibility55/100
Reproducibility scores 55, the mean over 2 Evolve-scored papers, both rated medium readiness. The TabPFN-3.5 report is dinged for closed-source headline variants; the FP8 quantization study is rigorous but Blackwell-only, 3 seeds, with no code stated. Open artifacts are the clearest lever for raising this score.
Research summary
Benjamin Jäger is a machine-learning researcher whose visible work centers on tabular foundation models — the TabPFN line — spanning both new architectures and efficient inference. This profile is name-resolved from OpenAlex bibliographic data; no affiliation or ORCID is attached in the snapshot. The Evolve catalog holds two 2026 arXiv papers. The first, TabPFN-3.5: Technical Report, is first-authored and claims state-of-the-art on seven tabular benchmarks via new cell encodings and a doubled-width transformer; Evolve judges the evidence broad and convincing while noting the headline variants are closed-source. The second, Attention Quantization for Tabular Foundation Models, reports FP8 attention with up to 1.7x speedup on TabPFN and TabICL at negligible accuracy loss — rigorous, Evolve notes, but Blackwell-only, three seeds, and no stated code. Bibliometrics sketch an early, tightly scoped record: 4 works in the recent three-year window (all dated 2026), an h-index of 1 with 0 i10-indexed papers, 49 distinct coauthors, and a field split evenly between foundation-model work and efficient inference. Among the scored radar axes, collaboration (78) and reproducibility (55, averaged over 2 papers) stand out; output and impact lack sufficient data to score. The reproducibility caveat is concrete: closed-source headline variants and no stated code release.
Collaboration pattern
Jäger's visible record is strongly collaborative: 49 distinct coauthors in the snapshot, scoring the collaboration axis 78. The recurring circle reads like a dedicated tabular foundation-model team — Frank Hutter, Noah Hollmann, Jonas M. Kübler, Klemens Flöge, Léo Grinsztajn, Nick Erickson, Oscar Key, Felix Birkel, Kürşat Kaya, and Adèle Frankel — with the two catalogued papers pairing a first-author lead with a middle-author contribution.
Research keywords
Highlights
TabPFN-3.5: first-authored technical report claiming SOTA on seven tabular benchmarks
FP8 attention quantization: up to 1.7x speedup for tabular foundation models
Timeline
- 2026First-authored 'TabPFN-3.5: Technical Report' on arXiv, claiming state-of-the-art on seven tabular benchmarks via new cell encodings and doubled model width.
- 2026Co-authored 'Attention Quantization for Tabular Foundation Models', reporting FP8 attention kernels in Triton with up to 1.7x speedup on TabPFN and TabICL.
- 2026Snapshot closes with 4 works in the recent three-year window (all dated 2026), 49 distinct coauthors, and a reproducibility mean of 55 across 2 scored papers.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Identity is name-resolved from OpenAlex bibliographic data; no ORCID, affiliation, or country is in the snapshot, so records by the same name may be conflated.
- All bibliometrics (coauthor counts, topic clusters, counts by year, h-index/i10) come from the OpenAlex-derived snapshot and may lag the live record.
- The reproducibility score (55) is the mean over only 2 Evolve-reviewed papers and may not represent the broader body of work.
- The timeline covers 2026 only, since the provided paper list contains no earlier years; no career or award history is included.