Huawei (China) machine-learning researcher spanning time series foundation models, forecasting, and AI for physics
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
13
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—
Not scored yet: no complete, verified per-publication record is available for this author, so career-wide output volume cannot be computed reliably. This axis would measure annual paper counts, venue tier, and authorship position across the full publication history.
Impact—
Not scored yet: citation coverage for this author is incomplete in the underlying data, so influence cannot be computed. This axis would measure citation uptake, h-index trajectory, and adoption of the open pretraining recipes beyond the original preprints.
Focus0/100
Scores 0, reflecting a field distribution split exactly evenly — one third each for time series forecasting, time series foundation models, and AI for physics / scientific multimodal learning. Topic counts nonetheless cluster: time series analysis and forecasting leads with 23, followed by domain adaptation and few-shot learning (12) and advanced graph neural networks (11), so the spread stays within one broader machine-learning program.
Collaboration62/100
Scores 62 on a 0-100 scale, built from 22 distinct coauthors in the snapshot. The recurring core includes Bahaeddine Abdessalem, Shifeng Xie, Zehao Xiao, Youssef Attia El Hili, Ambroise Odonnat, Jianfeng Zhang, Keli Zhang, Malik Tiomoko, Zhiwei Dong, Lei Zan, Themis Palpanas, and Henan Sun. On all three cataloged 2026 papers Pan holds a middle author position rather than first or corresponding, consistent with contributing inside larger lab teams.
Activity96/100
The strongest axis at 96. The snapshot records 26 works in the last three years and a counts-by-year curve rising from 1 work in 2012 and low single digits through the 2010s to 8 works each in 2022 and 2023, 6 in 2024, 7 in 2025, and 13 in 2026 — the highest annual count on record.
Reproducibility43/100
Mean readiness of 43 across the 3 papers Evolve assessed. Tabby and FlowTSFM rate medium — Tabby is singled out as an unusually transparent, well-ablated open pretraining recipe — while PhyMo rates low, with no code released. The pattern suggests artifacts accompany the time series foundation-model line more reliably than the AI-for-physics work.
Research summary
Lujia Pan is a researcher at Huawei Technologies (China) whose work divides evenly across three machine-learning subfields: time series forecasting, time series foundation models, and AI for physics with scientific multimodal learning, each holding one third of the field distribution in the OpenAlex-resolved snapshot. The three cataloged papers, all 2026 arXiv preprints, concentrate on foundation-model-scale sequence modeling. Tabby (145M parameters) ships an open, unusually transparent and well-ablated pretraining recipe and performs strongly on in-distribution GIFT-Eval while sitting mid-table on out-of-distribution TIME. FlowTSFM (38.8M parameters) recasts encoder depth as quantile transport with competitive benchmark results that remain preliminary pending ablations. PhyMo proposes PDE-structured physical fields as a new modality for multimodal AI4Physics, including solar forecasting, with multi-dataset gains but thin margins and no code. Evolve assesses reproducibility readiness as medium, medium, and low respectively, for a mean of 43 across three papers. Activity is the standout dimension: 26 works in the last three years, with annual counts climbing from 1 in 2012 to 8 in 2022 and 2023, and a record 13 in 2026 (score 96). The snapshot reports h-index 13 and i10-index 15, though output and impact dimensions remain unscored for insufficient coverage.
Collaboration pattern
Pan's collaboration footprint covers 22 distinct coauthors, with a recurring core that includes Bahaeddine Abdessalem, Shifeng Xie, Zehao Xiao, Youssef Attia El Hili, Ambroise Odonnat, Jianfeng Zhang, Keli Zhang, Malik Tiomoko, Zhiwei Dong, Lei Zan, Themis Palpanas, and Henan Sun. On all three cataloged 2026 papers Pan is a middle-position, non-corresponding author, consistent with contributing inside larger industry lab teams rather than leading as first or corresponding author.
Research keywords
Highlights
Open pretraining recipe for a 145M time series foundation model
Encoder depth recast as quantile transport
Physical fields proposed as a new multimodal modality
Timeline
- 2012First year in the snapshot's counts-by-year record, with 1 recorded work; output stays at 1-2 works per year through most of the 2010s.
- 2021Output steps up to 4 recorded works, doubling the prior typical annual pace.
- 2022A step-change year with 8 recorded works, a level repeated in 2023 (8) and approached in 2024 (6) and 2025 (7).
- 2026Most productive year on record with 13 works; three cataloged arXiv preprints appear — Tabby, FlowTSFM, and PhyMo — spanning time series foundation models and AI for physics.
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 indicators (h-index 13, i10-index 15, counts by year, coauthor and topic counts) come from an OpenAlex-resolved identity snapshot; name disambiguation may merge or split records.
- Output and impact dimensions are unscored because underlying coverage is insufficient; the quoted citation-style metrics should not be read as verified career totals.
- All three cataloged papers are 2026 arXiv preprints with Evolve review verdicts; reproducibility readiness reflects Evolve assessment, not independent replication.
- Paper-level claims rest on the three cataloged papers only; other recorded works in the counts-by-year data are not individually reviewed here.