Ajou University ML researcher: from few-shot learning and vision AI to efficient LLM compression and quantization
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
15
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 current snapshot lacks the venue-level publication metadata this axis needs. It would measure total scholarly output and its distribution across venues and years, beyond the coarse counts-by-year series already shown.
Impact—
Not yet scored. Although an h-index of 16 and an i10-index of 21 are recorded, per-paper citation detail is insufficient to normalize influence by field and year. This axis would measure citation impact relative to the discipline.
Focus0/100
Scores a perfect 100/100: the topic distribution is tightly concentrated on one arc. Domain Adaptation and Few-Shot Learning leads with 13 works, followed by Advanced Neural Network Applications (10) and COVID-19 diagnosis using AI (10), with Image and Video Retrieval (7) and Video Surveillance and Tracking (5); the 2026 catalog papers carry that focus into LLM compression and quantization.
Collaboration41/100
Scores 32/100, the narrowest axis, and the raw count explains why: only 4 distinct coauthors appear in the snapshot. Eunju Shin, Donggeon Lee, Dooyeon Na, and Seungmin Oh recur most often, and both 2026 catalog papers list Ryu as last author rather than corresponding, pointing to a small, stable senior-supervisor team pattern.
Activity90/100
Scores 86/100 on strong recent tempo: 18 works in the last three years, with counts by year rising from 3 in 2022 to a series peak of 9 in 2023, then 8 in 2024, 4 in 2025, and 6 in 2026 — a marked acceleration over the sparse early record of 1 work in 2007 and 1 in 2015.
Reproducibility66/100
Scores 55/100, but from a narrow base: reproducibility has been assessed for only 1 paper (reproMean 55), so the figure is provisional. Both catalog entries rate medium readiness — the LLM compression paper is flagged for borrowed baselines and thin statistics, while Colla-Q is well-ablated with strong calibration robustness.
Research summary
Jongbin Ryu (Ajou University, South Korea; ORCID 0000-0001-5574-5358) is a machine-learning researcher whose recorded output spans computer vision and, most recently, efficient large language models. The OpenAlex-linked snapshot shows a highly focused profile (focus 100/100) organized around five recurring topic clusters: Domain Adaptation and Few-Shot Learning (13 works), Advanced Neural Network Applications (10), COVID-19 diagnosis using AI (10), Advanced Image and Video Retrieval (7), and Video Surveillance and Tracking (5). Activity is strong (86/100), with 18 works in the last three years and a single-year peak of 9 in 2023, following a sparse early record (1 in 2007, 1 in 2015). An h-index of 16 and an i10-index of 21 are recorded, though per-paper citation detail is too thin to score the impact axis. Collaboration is the narrowest dimension (32/100): 4 distinct coauthors appear, with Eunju Shin, Donggeon Lee, Dooyeon Na, and Seungmin Oh the most frequent. The two Evolve-reviewed 2026 papers — on layer-wise curriculum learning for LLM compression and on Colla-Q, a MoE quantization method via minimax precision balancing — place Ryu in the last author position on both, consistent with a senior, supervisory role. Reproducibility is provisionally scored 55/100 from a single assessed paper; both catalog entries carry medium readiness ratings.
Collaboration pattern
Collaboration is the narrowest axis at 32/100, backed by just 4 distinct coauthors in the snapshot. Eunju Shin, Donggeon Lee, Dooyeon Na, and Seungmin Oh recur most often, and both 2026 catalog papers list Ryu as last author — not corresponding — a pattern consistent with a senior advisor overseeing a small, stable team rather than large international consortia.
Research keywords
Highlights
Colla-Q: minimax precision balancing for MoE quantization
Layer-wise curriculum learning for efficient LLM compression
A vision-first record converging on efficient LLMs
The five-topic snapshot — Domain Adaptation and Few-Shot Learning (13), Advanced Neural Network Applications (10), COVID-19 AI diagnosis (10), Image and Video Retrieval (7), Surveillance and Tracking (5) — traces a computer-vision core now redirecting toward LLM efficiency.
Timeline
- 2007Earliest year in the counts-by-year record: 1 indexed work appears, followed by a long gap until the series resumes in 2015.
- 20174 works — the densest year of the early phase of the record.
- 2020Output reaches 6 works, the highest point of the 2015–2022 stretch before the later surge.
- 20239 works — the single-year peak of the entire 2007–2026 series.
- 20248 works, sustaining the high tempo; the last three years total 18 works.
- 2026Two Evolve-reviewed arXiv papers list Ryu as last author — layer-wise curriculum learning for LLM compression and Colla-Q MoE quantization; 6 works recorded in total for the year.
Papers in the Evolve catalog
Layer-wise Curriculum Learning for Efficient LLM Compression
Colla-Q: Toward Collaborative Experts in MoE Quantization via Minimax Precision Balancing
Train Overcomplete, Deploy Compact: Scaling Recovery Capacity for Structured LLM Pruning
Re-calibrated Contrastive Loss for Transformation-Aware Prompt Conditioning in Vision-Language Models
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 (h-index, i10-index, counts by year, topics, distinct coauthors) come from an OpenAlex-linked snapshot; the profile is name-resolved from public bibliographic data, not a verified biography.
- The output and impact radar axes are unscored because venue and per-paper citation data are insufficient in the current snapshot, even though h-index 16 and i10-index 21 are recorded.
- Reproducibility (55/100) rests on a single assessed paper, so it is a provisional score rather than a stable estimate.
- Only 2 Evolve-reviewed papers (both 2026 arXiv preprints) inform the qualitative assessments; both list the researcher as last author and not corresponding.