ML researcher in tabular foundation models, in-context learning, and self-supervised physiological time-series
3
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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—
Output is not yet scored: publication-volume data is insufficient for a profile-level productivity estimate. This axis would measure total and recent output, venue spread, and first/corresponding-author share across the full publication record.
Impact—
Impact is not yet scored: citation data is too sparse to characterize influence, with only early-stage OpenAlex counts available. This axis would measure cumulative citations, citation velocity, and uptake of the 2026 papers.
Focus—
Focus is not yet scored: OpenAlex topics mix the 2026 machine-learning work with unrelated clusters such as hydrocarbon exploration, so concentration is unresolved. This axis would measure thematic concentration versus diversification.
Collaboration48/100
Evolve scores collaboration at 48. The OpenAlex record lists 10 distinct coauthors, with recurring partners including Tian Zhou, Bingqing Peng, Wenwei Wang, and Linxiao Yang. All three 2026 catalog papers are middle-author, non-corresponding contributions, consistent with team-based deep-learning research rather than solo or lead-author work.
Activity40/100
Evolve scores activity at 40, based on a sparse but recent record: recent3yWorks is 3, and OpenAlex countsByYear shows 2 works in 2023 and 3 in 2026, with earlier isolated entries in 1997 and 2003. The 2026 cluster — three arXiv papers in the Evolve catalog — indicates an active current phase.
Reproducibility—
Reproducibility is not yet scored at profile level: systematic code- and data-availability evidence is insufficient. This axis would measure released code, pretrained weights, and independent replication; Evolve's per-paper reviews rate the three 2026 papers at medium readiness.
Research summary
Mengni Ye is a machine learning researcher whose Evolve-reviewed output centers on tabular foundation models and self-supervised representation learning for physiological signals. Three 2026 arXiv papers, all with middle-author positions, form connected threads. One derives an FFN-free, attention-gated in-context learning model that Evolve reports beats TabPFN-3 by 31.4 Elo on TabArena, backed by strong intervention evidence. A second, Support-Compiled Feature Folding, folds wide tables into bounded leaves, giving accuracy and NLL gains across six backbones with 2.1–2.4x memory savings, though benefits hinge on feature selection. The third, Transferable Evidence Reconstruction, applies cross-group reader-transfer self-supervised learning to continuous glucose monitoring, with controlled gains across 14 tasks while private pretraining data tempers SOTA claims. OpenAlex bibliometrics remain thin: h-index 2, i10-index 1, sparse yearly counts, and a collaboration score of 48 versus 40 for activity on Evolve's radar. The topic profile mixes neural-network themes with older clusters such as hydrocarbon exploration, suggesting possible name-adjacent records. All three papers carry medium reproducibility readiness, and the profile is name-resolved from public bibliographic data.
Collaboration pattern
Ye's visible footprint is collaborative rather than solo: the OpenAlex record lists 10 distinct coauthors, and the frequent-coauthor list includes Tian Zhou, Bingqing Peng, Wenwei Wang, Beverly Jin, Jinjie Gu, Liang Sun, Linxiao Yang, Xue Wang, and Zuyi Zhu. All three 2026 catalog papers are middle-author, non-corresponding contributions, consistent with a contributor role in a deep-learning team; Evolve's collaboration score is 48.
Research keywords
Highlights
FFN-free attention-gated model for tabular in-context learning
Memory-lean feature folding across tabular foundation models
Transferable self-supervised learning for glucose time series
Timeline
- 2026Middle-author co-author of 'What Do Tabular Foundation Models Compute In Context?', deriving an FFN-free, attention-gated tabular ICL model; Evolve reports a 31.4 Elo gain over TabPFN-3 on TabArena.
- 2026Middle-author co-author of 'Transferable Evidence Reconstruction for Longitudinal Glucose Representations', applying cross-group reader-transfer self-supervised learning to continuous glucose monitoring.
- 2026Middle-author co-author of 'Support-Compiled Feature Folding', folding wide tables into bounded leaves for 2.1–2.4x memory savings across tabular foundation models.
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.
- Bibliometric signals are drawn from OpenAlex only; institutional affiliation and country are not available in the source data.
- The citation record is thin (h-index 2, i10-index 1) and yearly counts include 1997 and 2003 entries, so this OpenAlex identity may aggregate name-adjacent authors; the profile is name-resolved from public bibliographic data.
- OpenAlex topic clusters include hydrocarbon exploration, reservoir engineering, and stock-market forecasting, which do not match the 2026 machine-learning catalog papers; interpret topic-based fields with caution.
- Paper-level analysis covers only the three Evolve-reviewed 2026 arXiv papers; other works lack Evolve verdicts and were not assessed.