Multimedia security researcher at USTC — steganography and media forensics core, extending into generative-model safety
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
76
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—
The composite output score is not available in this snapshot. This axis would measure publication volume, productivity trends across the career span, and venue distribution over the full indexed record, beyond the raw counts-by-year figures already shown.
Impact—
No impact score has been computed yet. This axis would quantify citation-based influence — citation counts, field-weighted citation impact, and uptake of proposed methods — complementing the raw h-index (76) and i10-index (351) already present in the snapshot.
Focus100/100
Focus scores 100 because the topic profile is strongly concentrated: Advanced Steganography and Watermarking Techniques leads with 253 works and Digital Media Forensic Detection adds 209 — together about 58% of the 798 topic-tagged works — while Adversarial Robustness (135), Neural Network Applications (106), and GAN Image Synthesis (95) extend the same security-and-generation core, and the field distribution lists a single field.
Collaboration53/100
The collaboration score is 53, resting on 13 distinct co-authors recorded in the recent three-year window that covers 217 works — a compact, stable circle relative to that volume. Recurring names include Qi Chu, Xinghao Wang, Dong Li, Wei Yu, Yingwei Pan, Tao Gong, and Ting Yao. On the three Evolve-reviewed 2026 papers the researcher sits in middle or last author positions, consistent with senior-group collaboration rather than solo output.
Activity100/100
Activity scores 100 on a sustained and accelerating record: indexed works climb from 49 in 2018 and 55 in 2019 to 71 in 2023, 73 in 2024, and a peak of 81 in 2025, with 63 already logged for 2026. The last three years alone account for 217 works, and coverage reaches back to a single entry in 1992, spanning more than three decades of continuous publishing.
Reproducibility55/100
The reproducibility score of 55 is computed from only one paper with reproducibility data (reproMean 55), so the estimate is thin. All three Evolve-reviewed 2026 papers grade medium reproReadiness: EraseSAE reports single-run metrics, TimeSteer is validated mainly on its own benchmark, and the LLM safeguard framework relies on LLM-channel coding counts and a small depth stratum.
Research summary
Nenghai Yu is a researcher at the University of Science and Technology of China whose indexed record, name-resolved from OpenAlex bibliographic data, centers on information hiding and multimedia security. The snapshot lists an h-index of 76 and an i10-index of 351, with topic tags led by Advanced Steganography and Watermarking Techniques (253 works) and Digital Media Forensic Detection (209), followed by Adversarial Robustness in Machine Learning (135), Advanced Neural Network Applications (106), and GAN-based Image Synthesis (95). Output has accelerated markedly: from 26 indexed works in 2010 and 37 in 2017 to 71 in 2023, 73 in 2024, and a peak of 81 in 2025, with 217 works in the last three years. The three papers Evolve has reviewed — all 2026 arXiv preprints — show the record extending into generative-model and AI-safety territory: EraseSAE on sparse-autoencoder concept erasure in text-to-video diffusion, TimeSteer on training-free speech scheduling in joint audio-visual diffusion, and a framework converting LLM safeguard metrics into deployment-safety bounds. Evolve grades all three medium for reproducibility readiness, citing single-run metrics and benchmark-specific validation. With a collaboration score of 53 built on 13 distinct recent co-authors, the profile suggests a senior, high-throughput group leader whose focus is broadening from classical steganography and forensics toward the safety of modern generative systems.
Collaboration pattern
Within a recent three-year window of 217 indexed works, the snapshot records only 13 distinct co-authors, indicating a compact, stable team rather than loose large-scale networking; the collaboration score is 53. Recurring names include Qi Chu, Xinghao Wang, Dong Li, Wei Yu, Yingwei Pan, Tao Gong, and Ting Yao. On the three Evolve-reviewed 2026 papers the researcher sits in middle or last author positions, consistent with supervising contributions from a home group.
Research keywords
Highlights
Surgical concept erasure in text-to-video diffusion via sparse autoencoders
Training-free speech scheduling in joint audio-visual diffusion
Turning LLM safeguard metrics into deployment-safety bounds
Timeline
- 1992Earliest entry in the indexed counts-by-year record appears, with a single publication.
- 2017Indexed output reaches 37 works, part of a steady climb from 26 in 2010 through the 2010s.
- 2023Annual output reaches 71 indexed works and continues rising to 73 in 2024.
- 2025Output peaks at 81 indexed works; the recent three-year window totals 217 works.
- 2026Three arXiv preprints are reviewed by Evolve — EraseSAE, TimeSteer, and an LLM safeguard evaluation framework — with 63 works already indexed for the year.
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 are computed from an OpenAlex snapshot and reflect only indexed works.
- Output and impact dimension scores are not yet computed; only the raw h-index (76) and i10-index (351) are available.
- The reproducibility score (55) is based on a single paper with reproducibility data.
- Only three 2026 arXiv papers carry individual Evolve reviews; the rest of the record is summarized bibliometrically.