Korea University researcher: efficient document OCR via diffusion decoding and VLM-driven compositional SVG generation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
17
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
No composite output score is available yet — the snapshot lacks the aggregation needed to compute one. This axis would measure total and per-year publication volume; the yearly counts on file are raw signals only, not a score.
Impact—
A composite impact score is not yet computed; this axis would measure citation-based influence such as per-paper citations and field-weighted impact. Only raw h-index (17) and i10-index (21) figures are on file, and no score is derived from them here.
Focus0/100
The focus score of 0 reflects breadth rather than single-line depth: the field distribution splits exactly 50/50 between machine learning for document understanding and efficient inference (OCR vision-language models) and computer vision for SVG/vector-graphics generation. Topic counts reinforce this spread, led by multimodal machine learning applications (28) and advanced neural network applications (24), with three smaller clusters (9, 8, 8).
Collaboration50/100
The collaboration score of 50 sits at the midpoint, consistent with a compact network: 11 distinct coauthors appear in the snapshot, and the frequent-coauthor list — including Dohyun Kim, Sungjun Han and Seung Wook Kim — suggests a stable pool of repeat collaborators rather than large, rotating consortium-style author groups.
Activity100/100
Activity is the strongest axis at 100. The record shows 30 works in the most recent three-year window (2024–2026), with yearly counts climbing from 5 (2024) to a peak of 15 (2025) and 10 already logged in 2026 — an accelerating output cadence spanning both research lines.
Reproducibility78/100
Reproducibility averages 78 across the two Evolve-reviewed papers. The document-OCR paper rates high — a 3.94× decode speedup at −0.32 Overall with rigorous ablations, though on a single base model without confidence intervals — while the compositional SVG pipeline rates medium, its claimed upper bound being only greedily approximated on its own benchmark.
Research summary
Paul Hongsuck Seo is a machine-learning researcher at Korea University whose profile is name-resolved from OpenAlex bibliographic data. His recorded output splits evenly (50/50) between two lines: document understanding and efficient inference with OCR-oriented vision-language models, and computer vision work on SVG and vector-graphics generation. Around that core, topic counts cluster in multimodal machine learning applications (28) and advanced neural network applications (24), with smaller clusters in topic modeling (9), image and video retrieval (8), and GAN-based image synthesis (8). The bibliometric record spans 2014–2026 and shows accelerating activity: yearly counts rise from 5 (2024) to 15 (2025), with 10 already logged in 2026, summing to 30 works over the recent three-year window — the basis for the maximum activity score of 100. Citation signals on file include an h-index of 17 and an i10-index of 21, although composite output and impact scores are not yet computed. Evolve has reviewed two of his 2026 arXiv papers: GravityOCR, reporting a 3.94× decode speedup for document OCR at −0.32 Overall with high reproducibility readiness, and a VLM-driven compositional SVG pipeline with medium readiness. Averaged reproducibility across the two reviews is 78.
Collaboration pattern
The snapshot records 11 distinct coauthors and a recurring core — Dohyun Kim, Sungjun Han, Hyungguk Kim, Yusik Kim, Jamin Shin, Hongjoon Ahn, Sehwan Park, Taehoon Kim, Geonhee Han and Seung Wook Kim — consistent with a stable, mid-sized collaboration pool (score 50). Both reviewed 2026 papers are multi-author arXiv works, with Seo in a middle author position on the OCR paper and last author on the SVG paper; neither is marked corresponding.
Research keywords
Highlights
GravityOCR: 3.94× faster document OCR via self-speculative decoding
Compositional SVG generation via VLM-driven hierarchical parsing
2025–2026 output surge drives a perfect activity score
Yearly counts jump from 5 works (2024) to 15 (2025), with 10 already logged in 2026 — 30 works over the recent three-year window, the raw basis for the activity score of 100.
Timeline
- 2014Earliest year in the bibliometric record, with 1 work logged.
- 20163 works logged — the first visible ramp-up in the publication record.
- 2022Output reaches 5 works in a single year.
- 2025Peak activity: 15 works, the highest annual count in the record.
- 202610 works logged so far, including both Evolve-reviewed arXiv papers on diffusion-accelerated OCR and compositional SVG generation.
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.
- Bibliometrics come from an OpenAlex snapshot; the profile is name-resolved from public bibliographic data rather than biographically verified.
- Composite output and impact scores are not computed (marked missing); only raw h-index 17, i10-index 21 and yearly counts are on file.
- The reproducibility score of 78 averages only two Evolve-reviewed papers, both 2026 arXiv preprints.
- 2026 publication counts are partial-year and may still rise.