IBM researcher in adversarial ML: audio deepfake detection, LLM-agent robustness, and self-improving red-teaming.
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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 output axis is unscored: the captured OpenAlex record lacks reliable aggregate publication-volume metrics for this identity, beyond a thin recent-work slice. It would measure cumulative work counts, venue spread, and yearly productivity trends.
Impact—
The impact axis is unscored: citation coverage for this identity is thin (recorded h-index 1 and i10 index 0 against 11 recent works), so no reliable citation-based impact can be computed. It would measure citations, h-index, and scholarly influence.
Focus100/100
Focus scores 100, the maximum: the assigned field distribution contains a single field — speech and audio security (audio deepfake detection) — and the topic profile is dominated by adversarial robustness in machine learning (9 topic hits), ahead of topic modeling and AI ethics (5 each). Both catalog papers sit on the same agentic-robustness line of work, indicating a tightly concentrated research agenda.
Collaboration44/100
Collaboration scores 44: the snapshot records 8 distinct coauthors, a moderate network size. Recurring names include Xiang Li (listed twice), Wenqi Wei, Chen Xiong, Zhiyuan He, Stjepan Picek, and Tsung-Yi Ho. On both catalog papers the researcher holds a middle authorship position and is not the corresponding author, consistent with contributing inside larger teams rather than leading as sole or senior author.
Activity72/100
Activity scores 72: 11 works fall in the recent three-year window, and counts by year show a sharp recent ramp — 1 work in 2024, 1 in 2025, then 9 in 2026, the year of both Evolve-catalog papers. The profile is therefore early-weighted and front-loaded: nearly all captured output is concentrated in the most recent year.
Reproducibility55/100
Reproducibility scores 55, a mean over just 1 scored paper. The single scored item, the ROGUE paper on adversarially trained workflow generation for audio deepfake detection, carries Evolve's 'medium' reproducibility readiness. With only one paper assessed and the second catalog paper not yet scored, this axis rests on a thin evidence base and should be read as provisional.
Research summary
Pin-Yu Chen is a researcher at IBM (United States); this profile is name-resolved from public OpenAlex bibliographic records. The captured topic profile centers on adversarial robustness in machine learning (9 topic hits), supported by topic modeling (5) and ethics and social impacts of AI (5), with smaller clusters in natural language processing (2) and multi-agent systems and negotiation (2). The reviewed record concentrates in 2026: a middle-author arXiv paper, Robust Workflow Generation via Adversarial Learning for Audio Deepfake Detection, presents ROGUE, an adversarially trained LLM agent that orchestrates deepfake detectors — Evolve finds the robustness gains solid but notes that superiority over top single detectors is not shown, with medium reproducibility readiness — and a second 2026 arXiv paper, SIR: Self-improving Red-teaming for Compute Use Agents, extends the agenda to red-teaming compute-use agents. Activity is front-loaded: counts by year run 1 (2024), 1 (2025), and 9 (2026), totalling 11 works in the recent three-year window (activity score 72). Focus scores 100 because the field distribution lists a single field, speech and audio security (audio deepfake detection). Collaboration is moderate (score 44; 8 distinct coauthors recorded). Output and impact are not scored, since citation capture is thin (h-index 1, i10 index 0).
Collaboration pattern
The snapshot records 8 distinct coauthors (collaboration score 44). Recurring names include Xiang Li (listed twice, suggesting repeated collaboration), Wenqi Wei, Chen Xiong, Zhiyuan He, Stjepan Picek, and Tsung-Yi Ho. On both Evolve-catalog papers the researcher appears in a middle authorship position and is not the corresponding author, indicating contribution within team projects rather than sole leadership. Coauthor coverage likely reflects only the captured slice of output.
Research keywords
Highlights
ROGUE: adversarially trained LLM-agent workflows harden audio deepfake detection
SIR: self-improving red-teaming for compute-use agents
Second 2026 arXiv paper, also middle authorship, extending the agentic-robustness agenda toward self-improving red-teaming of compute-use agents. It is not yet fully reviewed in the Evolve catalog — no verdict, keywords, field, or repro-readiness data are available — so it is listed here without assessment.
A sharp 2026 output ramp across adversarial-ML topics
Counts by year show 1 work in 2024 and 1 in 2025, then 9 in 2026 — 11 works in the recent three-year window. Topic hits concentrate on adversarial robustness in machine learning (9), with topic modeling and AI ethics at 5 each, confirming a recently intensified, security-oriented research agenda.
Timeline
- 2024First work recorded in the OpenAlex snapshot (1 work); the captured record is recent-weighted.
- 2025One recorded work, immediately before the 2026 output surge.
- 2026Nine recorded works, including both Evolve-catalog papers: ROGUE (adversarial workflow generation for audio deepfake detection) and SIR (self-improving red-teaming for compute-use agents).
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 scores are computed from OpenAlex records only; the identity is name-resolved from public bibliographic data, not from an ORCID (none provided).
- Output and impact axes are unscored because citation capture is thin (recorded h-index 1 and i10 index 0 against 11 recent works), so citation-based measures look unreliable.
- Only two papers from this researcher are in the Evolve catalog, and only one has a full review (verdict, keywords, repro-readiness); the reproducibility score therefore rests on a single paper.
- The name Pin-Yu Chen appears within the supplied co-author list itself, suggesting name-resolution noise in the underlying co-authorship graph; coauthor counts may be incomplete.