Yingda Shen
Chinese University of Hong Kong, Shenzhen · United States
Early-career researcher pairing audio language models with efficient LLM reasoning amid accelerating 2026 output.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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—
Output is not yet scored: venue and role metadata are too sparse in this snapshot. This axis would measure work volume, venue tier, and authorship share over time; raw counts show 12 works in the last three years.
Impact—
Impact is not yet scored: citation signals are too thin to evaluate — an h-index of 1 and i10-index of 0 reflect an early career, not influence. This axis would measure citations, yearly citation rates, and uptake of the 2026 papers.
Focus0/100
The focus score is 0, the lowest on this profile: recorded output splits exactly evenly across two distinct fields — audio machine learning (speech, sound, and music understanding) at 0.5 and machine learning for efficient LLM inference and reasoning at 0.5. Topic records nonetheless lean audio-centric, with speech recognition and synthesis (7) and speech and audio processing (6) dominating, suggesting a speech-and-audio core now extending into LLM reasoning rather than unstructured drift.
Collaboration55/100
Collaboration scores 55, a moderate middle range: OpenAlex records 15 distinct coauthors, with repeated ties concentrated on Yuxiang Wang (the most frequent partner) and Zhizheng Wu. Both cataloged 2026 papers list the researcher in a middle authorship position without corresponding-author status, indicating a contributor role within a small, stable team rather than a solo or lead-driven pattern.
Activity75/100
Activity scores 75, the strongest axis here: OpenAlex counts 2 works in 2024 and 1 in 2025, then 9 works in 2026 — a steep ramp that lifts the three-year total to 12 works. Both cataloged arXiv papers appeared in 2026, confirming that the acceleration is current and concentrated in audio language models and efficient LLM reasoning rather than spread across older projects.
Reproducibility38/100
Reproducibility scores 38, averaged over 2 reviewed papers. The two releases diverge: EvoAudio is rated low because its evaluation reports gains of up to 6.3 points across five large audio-language models over 13 rounds with no error bars and no code, while RecurTrace is rated medium. Systematic release of code, seeds, and confidence intervals would lift this score materially in future updates.
Research summary
Yingda Shen is an early-stage researcher at the Chinese University of Hong Kong, Shenzhen, whose recorded work splits evenly between two fields: audio machine learning (speech, sound, and music understanding) and machine learning for efficient LLM inference and reasoning (0.5 / 0.5 by field distribution). OpenAlex records 12 works over the past three years, with output accelerating sharply — 2 works in 2024, 1 in 2025, and 9 in 2026 — alongside an h-index of 1 and an i10-index of 0, consistent with a career in its first years. Topic records lean audio-centric: speech recognition and synthesis (7) and speech and audio processing (6) dominate, with smaller clusters in topic modeling (3), phonetics and phonology (2), and music and audio processing (2). Two cataloged 2026 arXiv papers frame the current agenda: EvoAudio applies recursive self-improvement with GRPO-based verifiable rewards to audio language models, while RecurTrace introduces loop-time memory and oracle-distilled halting for adaptive computation in looped transformers. Both list the researcher in a middle, non-corresponding authorship position, and collaboration spans 15 distinct coauthors. Reproducibility signals are mixed: one paper is rated low (no error bars or code) and one medium, yielding a reproducibility score of 38.
Collaboration pattern
Collaboration is moderately networked: 15 distinct coauthors, with Yuxiang Wang the most frequent partner (three listings) followed by Zhizheng Wu (two). The wider circle includes Shengbo Cai, Ming-Hao Hsu, Qinke Ni, Liqiang Zhang, Teddy Sun, Steve Yevs, and Kunyu Feng. On both cataloged 2026 papers the researcher holds a middle, non-corresponding position, suggesting contribution within a team rather than independent leadership so far.
Research keywords
Highlights
Recursive self-improvement for audio understanding
Adaptive compute for looped transformers
A 2026 dual-track research surge
OpenAlex records 9 works in 2026 versus 3 across 2024 and 2025 combined. The two cataloged papers sit one in each field, pairing audio-model self-improvement with latent-reasoning efficiency — an early-career portfolio built around recursive and adaptive computation.
Timeline
- 2024OpenAlex records 2 works, the earliest activity captured in this snapshot.
- 2025A quiet year with only 1 recorded work.
- 2026Output accelerates to 9 works — the highest single-year count — lifting the three-year total to 12.
- 2026EvoAudio appears on arXiv, bringing recursive self-improvement with GRPO-based verifiable rewards to audio language models (middle authorship).
- 2026RecurTrace appears on arXiv, proposing loop-time memory and oracle-distilled halting for adaptive latent reasoning (middle authorship).
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 metrics are derived from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may merge or omit same-name records.
- Citation-impact data are too sparse to score (h-index 1, i10-index 0), so the output and impact dimensions are reported as missing rather than estimated.
- Only 2 papers have been reviewed in the Evolve catalog; unreviewed works may change the research-focus and reproducibility picture.
- The reproducibility score averages just 2 papers and reflects review-time code and statistics reporting, not independent replication.