IBM researcher spanning optimization, adversarial robustness, watermarking, and Koopman-based time-series forecasting.
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
11
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
This axis would measure annual publication volume and productivity trends, but the underlying per-work data is not yet available for scoring; the counts-by-year series (3 works in 2016 to 24 in 2026) is suggestive, not scored.
Impact—
This axis would measure citation impact — citations per work, highly-cited papers, and field-normalized influence — but such data is insufficient here; although the snapshot lists h-index 11 and i10-index 14, no citation distribution is available for scoring.
Focus100/100
Focus scores a perfect 100, reflecting a profile concentrated in machine learning: the OpenAlex field distribution lists a single field — machine learning (time-series forecasting / dynamical systems) — while the top topic clusters (Domain Adaptation and Few-Shot Learning, Stochastic Gradient Optimization Techniques, and Adversarial Robustness in Machine Learning, 20 works each) stay within core ML, with steganography and watermarking (15) as a security-leaning satellite.
Collaboration54/100
Collaboration scores 54, based on 14 distinct coauthors in the record. The pattern points to a senior, hands-on role: all three cataloged 2026 arXiv papers list Bu in the last author position. Recurring partners include Yepeng Liu (listed twice among frequent coauthors), Ruiquan Li, Tianyi Chen, Xuandong Zhao, Dawn Song, Dasen Dai, Chengzhi Liu, Yiren Song, Hai Ci, Yu Zhang, and Qi Zhang — a network spanning optimization, security, and generative-AI groups.
Activity100/100
Activity tops the scale at 100. The counts-by-year series shows sustained and accelerating output: 3 works in 2016, 6 in 2019, 8 in 2022, 13 in 2024, 16 in 2025, and 24 in 2026 — the highest annual total on record — with 53 works in the last three years. This pace, alongside three 2026 arXiv preprints in Evolve's catalog, marks an unusually productive current phase.
Reproducibility55/100
Reproducibility scores 55, computed from a mean of 55 over one assessed paper (reproPapers: 1), so the estimate is thin. Consistent with this, the two cataloged 2026 preprints carry a 'medium' repro-readiness rating: K2SVD's forecasting results are fast and competitive but evaluated with point metrics only, and FigTree tops PaperBananaBench yet remains judge-dependent and statistically untested. More assessed papers would firm this axis up.
Research summary
Yuheng Bu is a machine learning researcher at IBM in the United States whose OpenAlex-resolved profile spans stochastic gradient optimization, domain adaptation and few-shot learning, adversarial robustness, and steganography and watermarking, extending recently into dynamical-systems approaches to time-series forecasting. The record shows an h-index of 11 and an i10-index of 14, with output accelerating steadily: 3 indexed works in 2016, 6 in 2019, 13 in 2024, 16 in 2025, and 24 in 2026, totalling 53 works over the past three years. Evolve's catalog includes three 2026 arXiv preprints with Bu as last author. The Koopman paper (K2SVD) pairs principled low-rank Koopman learning with Kalman filtering and is judged genuinely fast and competitive at forecasting, though its evaluation is point-metric-only despite probabilistic framing. FigTree recasts scientific figure generation as recursive SVG program synthesis with multi-agent LLMs, topping PaperBananaBench while remaining judge-dependent and statistically untested. A third preprint studies learning to follow in-context watermark instructions via self-distillation. The overall pattern is a productive, broad-ranging researcher who increasingly anchors projects as senior collaborator.
Collaboration pattern
The record shows 14 distinct coauthors and a clear senior-author pattern: on all three cataloged 2026 preprints Bu is the last author (non-corresponding). Frequent partners include Yepeng Liu — the only name listed twice — alongside Ruiquan Li, Tianyi Chen, Xuandong Zhao, Dawn Song, Dasen Dai, Chengzhi Liu, Yiren Song, Hai Ci, Yu Zhang, and Qi Zhang, suggesting steady joint work across optimization, model security, and generative-AI teams.
Research keywords
Highlights
K2SVD: principled Koopman representations for fast time-series prediction
FigTree: scientific figures as recursive, editable SVG programs
Learning to follow in-context watermark instructions via self-distillation
A third 2026 arXiv preprint with Bu in the last author position; it connects to the watermarking and steganography thread in the topic record (15 works on advanced steganography and watermarking techniques). Evolve has not yet published a verdict, keywords, or reproducibility rating for this paper.
Timeline
- 2016Earliest indexed works appear: 3 publications recorded in the OpenAlex counts-by-year series.
- 2019Output edges up to 6 works, an early sign of steady growth.
- 20228 works indexed, continuing a gradual climb through the early 2020s.
- 2024Output accelerates to 13 works in a single year.
- 202516 works indexed as the upward trend steepens.
- 2026Peak year so far: 24 works, including three arXiv preprints in Evolve's catalog (Koopman forecasting, scientific figure generation, and watermark self-distillation), all with Bu as last author.
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 data comes from OpenAlex (identity: openalex); the profile is name-resolved from public bibliographic records, and 'Yuheng Bu' may correspond to more than one researcher.
- The output and impact radar axes are unscored because per-work citation data is insufficient; h-index (11) and i10-index (14) come from the snapshot alone.
- The reproducibility score (55) is averaged over a single assessed paper (reproPapers: 1), so it should be read as preliminary.
- Evolve has reviewed two of the three cataloged papers; the watermark self-distillation preprint has no verdict, keywords, or reproducibility rating yet.