Audio deepfake detection and efficient LLM reasoning researcher at CUHK-Shenzhen (early-career, arXiv-first record)
2
papers here
2
reviewed by Evolve
1
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—
Not yet scored. The output axis would measure publication volume, venue quality, and year-over-year consistency; OpenAlex coverage for this arXiv-heavy record is incomplete, so the counts behind the 2023–2026 rise cannot yet support a reliable score.
Impact—
Not yet scored. Impact would measure citation uptake and influence beyond the author's own record; the snapshot's h-index of 2 and i10-index of 1 are too sparse to yield a reliable score at this career stage.
Focus0/100
The focus score is 0, matching the raw field distribution: an exact 50/50 split between audio deepfake detection (speech and audio signal processing) and machine learning for efficient LLM inference and reasoning. Topic counts — generative adversarial networks and image synthesis (7), speech recognition and synthesis (5), digital media forensic detection (5) — reveal shared threads of generative modeling and forensics that connect the two halves.
Collaboration46/100
Collaboration scores 46, computed from 9 distinct coauthors. The frequent-coauthor list is a compact, repeating core — Yuxiang Wang and Zhizheng Wu each appear twice, with Li Wang, Wan Lin, Yingda Shen, Haoning Xu, and Junyu Wang alongside — pointing to one or two stable working groups rather than a broad, dispersed network.
Activity84/100
Activity is the strongest axis at 84. The by-year record rises from 1 work in 2023 to 2 in 2024, 4 in 2025, and 11 in 2026, and 17 works fall within the last three years — a steep, current acceleration concentrated almost entirely in 2026.
Reproducibility55/100
Reproducibility averages 55 across 2 reviewed papers, both rated medium readiness. CoRELoop reports concrete, checkable numbers — pooled EER cut from 4.85% to 3.74% with strong ablations, though no external SOTA comparison; RecurTrace shows credible gains over fixed depth on MathQA, but its adaptive margin leans on a performance floor.
Research summary
Kunyu Feng is an early-stage researcher at the Chinese University of Hong Kong, Shenzhen (ORCID 0009-0004-8236-2781), profiled from public OpenAlex bibliographic data. The record splits evenly — a raw 50/50 field distribution — between audio deepfake detection within speech and audio signal processing, and machine learning aimed at efficient LLM inference and reasoning. Activity is the strongest measured axis at 84: the by-year work record climbs from 1 (2023) to 2 (2024), 4 (2025), and 11 (2026), with 17 works in the last three years. Topic counts reinforce this two-track program: generative adversarial networks and image synthesis (7), speech recognition and synthesis (5), digital media forensic detection (5), advanced neural network applications (4), and multimodal machine learning (3). Two Evolve-reviewed 2026 arXiv papers anchor the catalog. CoRELoop, first-authored, applies LoRA-based recurrent refinement to a frozen self-supervised speech detector, cutting pooled EER from 4.85% to 3.74%. RecurTrace contributes loop-time memory and oracle-distilled halting for looped transformers, with gains over fixed depth on MathQA. Collaboration scores 46 around 9 distinct coauthors, and reproducibility averages 55, with both reviewed papers rated medium readiness.
Collaboration pattern
Collaboration scores 46 from 9 distinct coauthors. The frequent-coauthor list shows a compact, repeating core: Yuxiang Wang and Zhizheng Wu each appear twice, alongside Li Wang, Wan Lin, Yingda Shen, Haoning Xu, and Junyu Wang. This pattern suggests work within one or two stable groups at CUHK-Shenzhen rather than a wide network, consistent with a record concentrated in 2026.
Research keywords
Highlights
CoRELoop: LoRA-based recurrent refinement cuts pooled EER from 4.85% to 3.74% in audio deepfake detection
RecurTrace: loop-time memory and oracle-distilled halting for adaptive latent reasoning
Timeline
- 2023First work appears in the OpenAlex by-year record (1 work), marking the start of the visible publication trail.
- 2024Output doubles to 2 works as the two research tracks begin to take shape.
- 20254 works appear, continuing steady growth before the 2026 acceleration.
- 2026Most productive year: 11 works, including both Evolve-reviewed arXiv papers — first-authored CoRELoop on audio deepfake detection and RecurTrace on adaptive latent reasoning.
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 figures come from an OpenAlex identity snapshot; arXiv-heavy records like this one may be under-counted or late-indexed.
- Output and impact axes are unscored due to insufficient data; the h-index of 2 and i10-index of 1 are too sparse to support inference.
- Only the 2 Evolve-reviewed papers carry qualitative assessments; other works in the by-year record are not independently reviewed.
- Profile is name-resolved from public bibliographic data; identity should be confirmed via ORCID 0009-0004-8236-2781.