Hefei University of Technology researcher in LLM unlearning, point cloud registration, industrial inspection
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
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—
Publication-volume data is incomplete in this snapshot: only three Evolve-reviewed arXiv preprints and a 22-work recent count are available, with no verified full publication list or venue breakdown, so total output cannot be scored yet.
Impact—
Citation impact cannot be scored: the recorded h-index (2) and i10-index (0) conflict with 22 recent works, suggesting incomplete identity aggregation. This axis would measure citations, h-index trajectory, and uptake of the reviewed preprints.
Focus0/100
The focus score of 0 reflects an unusually even spread rather than missing data: the field distribution splits exactly into thirds across LLM machine unlearning, 3D point cloud registration, and multimodal industrial inspection, and the five recorded topics (topic modeling and GAN image synthesis at 6 each, multimodal applications 5, human motion 4, NLP 3) cross at least three communities. Expect method reuse across problems rather than one deep vertical.
Collaboration60/100
Collaboration scores 60, drawn from 19 distinct coauthors in the recent window and 12 recurring partners, including Pengyang Shao, Chuanpeng Lu, Wei Qin, Yanzheng Jin, Xiaohao Liu, Xi Ai, Kenji Kawaguchi, Zhixin Cheng, Jiacheng Deng, Xiaotian Yin, Baoqun Yin, and Tianzhu Zhang. On the three reviewed papers Hong sits last-author twice and middle once — a senior, supervisory slot rather than hands-on first authorship.
Activity91/100
Activity is the strongest axis at 91: 22 works in the last three years, with the year-by-year counts recording all 22 in 2026 alone, and three separate arXiv preprints already reviewed by Evolve that year across unlearning, registration, and inspection. This cadence points to a continuously shipping group; the caveat is that 2026 is the only year with counts, so longer-term tempo is unverified.
Reproducibility43/100
Reproducibility averages 43 over three assessed papers. Evolve rates the PCBA VQA and multilingual unlearning papers medium, citing an unexplained evaluation gap and single-model evidence without significance tests respectively, and the dual-masked autoencoder paper low, citing citation errors and loose reinforcement-learning formalism. Empirical gains are often substantial, but reporting hygiene — baselines, statistics, references — lags the output volume.
Research summary
Richang Hong is a researcher at Hefei University of Technology (CN) whose recent work, as resolved from OpenAlex, spans three evenly weighted directions: LLM machine unlearning, 3D point cloud registration, and multimodal computer vision for industrial inspection — the field distribution assigns exactly one third to each. Activity is the standout signal: 22 works in the last three years, all 22 recorded in 2026 in the year-by-year counts, including three arXiv preprints reviewed by Evolve. The unlearning line contributes CLLPU, a ten-language benchmark separating insufficient from excessive unlearning propagation; the registration line proposes a dual-masked autoencoder with reinforcement-learned masking for image-to-point cloud registration; and the inspection line applies task-aware GRPO reward design to PCBA visual question answering, ranking 2nd (83.24) on the MM'26 challenge. Across all three reviewed papers Hong holds a senior slot — last author twice, middle author once — consistent with a supervisory role inside a 19-distinct-coauthor network. Citation-based measures warrant caution: the snapshot records h-index 2 and i10-index 0, hard to reconcile with 22 recent works and likely reflecting incomplete identity aggregation rather than actual impact, so the output and impact axes are left unscored. Reproducibility is moderate, averaging 43 across the three assessed papers.
Collaboration pattern
Collaboration is broad rather than deep: 19 distinct coauthors in the recent window and 12 recurring partners, including Pengyang Shao, Chuanpeng Lu, Wei Qin, Yanzheng Jin, Xiaohao Liu, Xi Ai, Kenji Kawaguchi, Zhixin Cheng, Jiacheng Deng, Xiaotian Yin, Baoqun Yin, and Tianzhu Zhang. On all three Evolve-reviewed papers Hong is not first author — last author twice, middle once — the typical posture of a supervising senior author whose junior collaborators lead the writing. Axis score: 60.
Research keywords
Highlights
CLLPU: a 10-language benchmark for multilingual LLM unlearning
Dual-masked autoencoder for image-to-point cloud registration
Task-aware GRPO rewards lift PCBA visual question answering
Timeline
- 2026Recent 3-year output reaches 22 works, all recorded in 2026 per OpenAlex year-by-year counts.
- 2026Released CLLPU, a 10-language benchmark probing propagation boundaries in multilingual LLM unlearning (arXiv, last author).
- 2026Proposed Mask 2D-3D, an adaptive dual-masked autoencoder for image-to-point cloud registration (arXiv, middle author).
- 2026Applied task-aware GRPO reward design to cross-domain PCBA VQA, ranking 2nd (83.24) on the MM'26 challenge (arXiv, 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.
- Bibliometrics come from OpenAlex only; no ORCID was available, so this profile is name-resolved from public bibliographic data and coverage of the researcher's record may be incomplete.
- Recorded h-index (2) and i10-index (0) conflict with 22 recent works, suggesting identity fragmentation; output and impact are left unscored rather than estimated.
- Qualitative judgments rest on three Evolve-reviewed papers, all 2026 arXiv preprints; no peer-reviewed venue information is in scope.
- Year-by-year counts cover 2026 only, so longer-term productivity trends cannot be assessed.