Henan Sun
The Hong Kong University of Science and Technology (Guangzhou)
First-author ML researcher at HKUST (Guangzhou) spanning LLM post-training RL and AI-for-physics multimodal learning.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 publication-output metrics are available yet — this axis would measure total paper count, venue tier, and yearly output volume. The current scope covers only two 2026 arXiv preprints reviewed by Evolve, both first-author.
Impact—
No impact data — citations, h-index, or usage signals — is available in this snapshot. This axis would measure citation uptake and scholarly influence; both cataloged preprints are 2026 works, too recent for meaningful citation records.
Focus0/100
The focus axis scores 0 on the 0-100 scale, and the raw field distribution explains why: the two cataloged papers split exactly 50/50 between LLM post-training/reinforcement learning and AI for physics/scientific multimodal learning. This is a two-front early profile — reward modeling and policy alignment on one side, PDE-structured physical-field modalities on the other — so the score reads as breadth across directions rather than a concentrated single-topic record.
Collaboration53/100
The collaboration axis scores 53 (0-100 scale) from a raw count of 13 distinct coauthors across the cataloged work. The frequent-coauthor list shows Haitao Hu, Jianfeng Zhang, and Nuo Chen appearing twice each, alongside Jia Li, Zehua Li, Qifan Zhang, Jin Liu, and Lujia Pan; the data does not map individual coauthors to specific papers, but the repeats point to at least one stable core group.
Activity—
No activity metrics — publication cadence, submission frequency, or recent-output trend — are present. This axis would measure publishing pace over time; the only dated evidence in scope is two arXiv preprints, both stamped 2026.
Reproducibility20/100
The reproducibility axis scores 20 (reproMean 20 across 2 papers). Evolve flagged both cataloged papers as low reproducibility readiness: the DCRL review cites thin theory and missing key ablations, and the PhyMo review notes thin margins, trivial theory, and no code release. Reproducing either result would currently require re-implementation from the papers alone.
Research summary
Henan Sun is a machine-learning researcher at The Hong Kong University of Science and Technology (Guangzhou), profiled here from public bibliographic data and two Evolve-reviewed papers. The current record is early-stage but coherent: both cataloged works are 2026 arXiv preprints with Sun as first author, and the raw field distribution splits evenly — 50/50 — between LLM post-training/reinforcement learning and AI for physics/scientific multimodal learning. DCRL decouples and couples reinforcement learning via policy-reward manifold alignment, co-evolving policy and reward through shared weights and evolving rubrics; Evolve's review highlights that a 4B model beats a 32B model while flagging thin theory and missing key ablations. PhyMo proposes PDE-structured physical fields as a new modality for multimodal AI4Physics, reporting promising multi-dataset gains that the review tempers as thin-margin, with trivial theory and no code release. Collaboration data shows 13 distinct coauthors and a collaboration score of 53, with Haitao Hu, Jianfeng Zhang, and Nuo Chen recurring in the frequent-coauthor list. Reproducibility is the weakest measured axis (20 across 2 papers), consistent with both reviews' low readiness flags. Output, impact, and activity axes are unscored because no citation or productivity data is in scope; the profile should be read as a directional snapshot of a first-author-heavy, two-front early research agenda rather than an established citation record.
Collaboration pattern
Sun's cataloged work lists 13 distinct coauthors and a collaboration score of 53 (0-100 scale). The frequent-coauthor list is anchored by repeat appearances from Haitao Hu, Jianfeng Zhang, and Nuo Chen, with Jia Li, Zehua Li, Qifan Zhang, Jin Liu, and Lujia Pan also recurring. Both papers are first-author (non-corresponding), suggesting a contributing-lead role within a team rather than solo or supervising-author work; the data does not map coauthors to specific papers.
Research keywords
Highlights
DCRL: policy and reward co-evolve — a 4B model beats 32B
PhyMo: physical fields proposed as a new multimodal modality for AI4Physics
A consistent early signature: ambitious reframings, low artifact readiness
Across both cataloged papers — both first-author 2026 arXiv preprints — Evolve flagged low reproducibility readiness, yielding a reproducibility mean of 20 over 2 papers. The pairing suggests a pattern of ambitious reframings (policy-reward manifold alignment; physical fields as a modality) that currently outrun released artifacts and ablation evidence.
Timeline
- 2026DCRL appears on arXiv: a first-author preprint on decoupling and coupling reinforcement learning via policy-reward manifold alignment, later reviewed by Evolve with a low reproducibility readiness flag.
- 2026PhyMo appears on arXiv: a first-author preprint proposing PDE-structured physical fields as a new modality for multimodal AI4Physics; Evolve notes promising but thin multi-dataset gains and no code.
- 2026With both 2026 preprints cataloged, the recorded field split stands at exactly 50/50 between LLM post-training/reinforcement learning and AI for physics/scientific multimodal learning, alongside 13 distinct coauthors.
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.
- Early-stage profile built from only two 2026 arXiv preprints; the data contains no peer-reviewed venues, citation counts, or ORCID linkage.
- Radar scores for output, impact, and activity are missing because the snapshot holds no publication-volume or citation data; they are reported as unavailable, not as zero.
- The reproducibility score (20) is a mean over just 2 papers, both flagged low readiness by Evolve reviews (missing ablations for DCRL; no code for PhyMo).
- "Henan Sun" is a plausible common name; this profile is name-resolved from public bibliographic data and identity is not independently verified beyond the reviewed papers.