Jian Huang
Hong Kong Polytechnic University · Hong Kong SAR China
Machine learning researcher at Hong Kong Polytechnic University: diffusion text generation and LLM privacy auditing.
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 usable output signal is available yet: OpenAlex lacks full publication counts and venue-normalized productivity beyond the recent window. This axis would measure total research output and its distribution across venues and years.
Impact—
Impact data is insufficient: citation counts, field-weighted citation impact, and h-index-based measures are not yet computable for this profile. This axis would measure citation-based influence relative to field and career stage.
Focus0/100
Focus scores 0 because OpenAlex field assignment splits evenly across three labels — machine learning (NLP), privacy auditing (stat.ML), and LLM membership inference — each about one-third. Topic counts echo this breadth: Topic Modeling, NLP Techniques, Privacy-Preserving Technologies, and Adversarial Robustness each appear twice. The strands actually connect around generative modeling and model auditing, so the low score reflects even spreading across adjacent subfields rather than scattered unrelated work.
Collaboration39/100
Collaboration scores 39, a modest value. OpenAlex records 6 distinct coauthors, and the catalog lists six frequent collaborators: Shengjie Niu, Yeheng Ge, Yangrong Hu, Ding Huang, Xueyu Zhou, Zebin Yun. All three reviewed papers place Huang in the last author position without a corresponding-author flag, consistent with a senior or supervisory role inside a small, stable team rather than a broad distributed network.
Activity61/100
Activity scores 61. OpenAlex counts 7 works in the last three years, with countsByYear placing all 7 in 2026 alone — a burst-like pattern typical of newly indexed, arXiv-first profiles. All three reviewed papers are 2026 arXiv preprints, so the mid-range score reflects genuine current momentum measured over a short observable window rather than a long publication record.
Reproducibility70/100
Reproducibility scores 70, the mean readiness over 3 reviewed papers. Per-paper readiness is medium for the RMDM diffusion paper and the pairwise likelihood-ratio membership inference paper, and high for the black-box word-level probability method (WPMIA). Evolve's verdicts still flag caveats — GPT-2-scale evidence, idealized theory, small-n evaluation — but the strongest artifact readiness sits on the black-box attack line.
Research summary
Jian Huang is a machine learning researcher at Hong Kong Polytechnic University whose OpenAlex record shows 7 works in the last three years, all dated 2026 — an early-stage profile with clearly active output. The reviewed work clusters on two connected fronts. In generative modeling, the Representation-based Masked Diffusion Model (RMDM) conditions masked diffusion on aligned text latents to improve few-step quality and throughput, drawing on flow matching, MeanFlow, and latent-variable modeling; Evolve notes the evidence is GPT-2-scale and baselines lack encoder knowledge. In language-model privacy auditing, two 2026 papers advance membership inference: one claims over 25% low-FPR TPR gains via Gaussian likelihood ratios, population calibration, and a Cauchy combination test, while the black-box WPMIA recovers likelihoods from sampled text and beats black-box baselines including proprietary LLMs. Across all three reviewed papers Huang holds the last author position, and the six reported frequent coauthors suggest a compact recurring team. The reproducibility dimension scores 70 (mean over 3 papers, readiness from medium to high), while field assignment splits roughly evenly across NLP, privacy auditing, and LLM membership inference, producing a focus score of 0. Citation-based output and impact axes remain unavailable; h-index and i10-index both stand at 0.
Collaboration pattern
OpenAlex records 6 distinct coauthors, and the catalog lists the same six as frequent collaborators — Shengjie Niu, Yeheng Ge, Yangrong Hu, Ding Huang, Xueyu Zhou, Zebin Yun — pointing to a compact, recurring team. In all three reviewed 2026 papers Huang occupies the last author position without the corresponding-author flag, a pattern consistent with supervisory authorship. The collaboration dimension scores 39, reflecting a small stable cluster rather than a broad international network.
Research keywords
Highlights
Few-step text generation via representation-conditioned masked diffusion (RMDM)
Stronger low-FPR membership inference with pairwise likelihood ratios
Black-box membership inference from sampled text (WPMIA)
Timeline
- 2026OpenAlex records 7 works in 2026 — the only year with recorded output in the recent three-year window, marking the profile's visible start.
- 2026Representation-based Masked Diffusion Model (RMDM) posted on arXiv; Huang is last author, with medium reproducibility readiness.
- 2026Membership Inference via Pairwise Likelihood Ratios posted on arXiv; last-author position, medium reproducibility readiness.
- 2026Black-Box Membership Inference via Word-Level Probability Estimation (WPMIA) posted on arXiv; last-author position and the profile's highest readiness rating (high).
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.
- Bibliometrics are drawn from an OpenAlex identity only; 'Jian Huang' is a common name, so this profile is name-resolved from public bibliographic data and may mix records of same-named authors.
- The output and impact axes are missing because citation-side data is unavailable; OpenAlex currently reports h-index 0 and i10-index 0, consistent with freshly posted 2026 arXiv preprints.
- Only the 3 Evolve-reviewed papers are individually assessed; the other works counted within the recent 7 are not reviewed here.
- The reproducibility score of 70 is the mean readiness across 3 reviewed papers per Evolve review, not an independent code or artifact audit.