NTU Singapore researcher in machine unlearning, GNN ownership verification and LLM knowledge-editing safety
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Not yet scoreable: the snapshot lacks a full works record and per-year output series, so total publication volume, venue mix and first-author share cannot be computed; only 27 recent-three-year works are visible.
Impact—
Not yet scoreable: beyond a minimal h-index of 2 and i10-index of 0, the snapshot carries no citation aggregates by year, field-normalized impact or citing-work network, so real influence on the field cannot be measured.
Focus0/100
The focus axis scores 0/100, mirroring the raw field distribution: the record splits exactly three ways across machine unlearning, GNN security and model IP protection, and LLM knowledge editing and safety (each 0.33). Topic counts likewise spread across five clusters — adversarial robustness (12), topic modeling (7), privacy-preserving technologies (7), multimodal applications (6) and GAN-based image synthesis (5) — indicating breadth rather than concentration.
Collaboration54/100
Collaboration scores 54/100. The snapshot lists 14 distinct coauthors, with recurring partners including Minyi Peng, Darian Gunamardi, Ivan Tjuawinata and Huaxiong Wang. On all three Evolve-reviewed papers Lam sits in the last author position, none as corresponding author — a supervisory signature; whether this holds across the wider 27-work record cannot be verified from the available data.
Activity97/100
Activity is the strongest axis at 97/100. The counts-by-year series records 27 works in 2026, equal to the 27 recent-three-year works — effectively all visible output sits in a single current burst. All three catalog-reviewed papers carry 2026 arXiv dates. Whether this rate can be sustained beyond one window is the open question the data cannot yet answer.
Reproducibility55/100
Reproducibility scores 55/100, computed as a mean across the three reviewed papers (reproPapers: 3), each rated medium readiness. Evolve's verdicts flag concrete gaps: accuracy-only single-run results in the GNN watermarking paper, privacy claims in tension with forget-class data usage in the label-removal method, and thin baselines with 1000-edit degradation in the edit-reversal work.
Research summary
Kwok-Yan Lam is a researcher at Nanyang Technological University whose OpenAlex record shows a concentrated burst of output: 27 works in the last three years, all dated 2026 in the snapshot, with an h-index of 2 and i10-index of 0 — a growth-stage profile. The reviewed work centers on trustworthy machine learning across three equally weighted fields: machine unlearning, GNN security and model intellectual-property protection, and NLP for LLM knowledge editing and safety. Evolve has reviewed three 2026 arXiv papers on which Lam is the last author (none corresponding): a watermark-based fingerprint framework for GNN ownership verification, an output-space redistribution method for adaptive label removal, and gated singular-vector shrinkage for reversing LLM knowledge edits. Each carries medium reproducibility readiness; the reproducibility mean across the three papers is 55/100. The radar scores activity 97/100 and collaboration 54/100 against 14 distinct coauthors, while output and impact remain unscored for lack of citation data. Topic signals extend to adversarial robustness in machine learning (12 records), topic modeling (7) and privacy-preserving technologies in data (7). This profile is name-resolved from public bibliographic data; ORCID is not linked, so identity should be confirmed before citing.
Collaboration pattern
The record lists 14 distinct coauthors, and the collaboration axis scores 54/100. Frequent partners include Minyi Peng, Darian Gunamardi, Ivan Tjuawinata, Yongsen Zheng, Han Zhang, Huaxiong Wang and Qianren Mao. A consistent pattern in the Evolve catalog is last-author placement on all three reviewed 2026 papers, none as corresponding author — a supervisory signature. Author roles across the remaining recent works are not visible in the snapshot.
Research keywords
Highlights
Watermark-based fingerprinting for GNN ownership verification
Near-zero-cost label removal via output space redistribution
Selective reversal of LLM knowledge edits
Timeline
- 2026Publishes on arXiv a robust watermark-based fingerprint framework for GNN ownership verification, as last author.
- 2026Publishes on arXiv OSR, an output-space redistribution method for adaptive label removal in classification models, as last author.
- 2026Publishes on arXiv selective knowledge edit reversal via gated singular vector shrinkage, as last author.
- 2026The OpenAlex snapshot records 27 works dated 2026 with 14 distinct coauthors — the visible peak of this record's output.
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.
- All bibliometrics come from a single OpenAlex snapshot; no citation-by-year series or venue-level data is included, which is why the output and impact axes are unscored.
- ORCID is not linked; this profile is resolved by name from public bibliographic data, so identity confirmation is recommended before citing.
- The three reviewed papers are 2026 arXiv preprints; Evolve verdicts note limitations such as single-run results and thin baselines, and none is corresponding-author work.
- Only 3 of the 27 recent works carry reproducibility assessments (reproMean 55), so that score covers a small slice of the record.