NAVER Cloud LLM post-training researcher (h-index 31): MoE routing, reasoning-trace efficiency, speculative decoding.
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
31
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 not yet scored: this snapshot lacks the publication-count aggregates needed to compute it. Once available, it would measure total works, first-authored share, and how output volume and venues evolve over time.
Impact—
The impact axis is not yet scored: citation aggregates were insufficient at snapshot time, though the raw record lists h-index 31 and i10-index 46. It would measure citation totals, recent citation velocity, and field-relative influence.
Focus0/100
The focus axis scores 0, reflecting breadth over concentration: the field distribution splits evenly (0.5/0.5) between two closely related LLM post-training labels, and OpenAlex topics spread across five clusters — Advanced Neural Network Applications (71), Domain Adaptation and Few-Shot Learning (54), Topic Modeling (29), Multimodal (26), and Data Classification (23).
Collaboration50/100
Collaboration scores 50. Over the recent three-year window, 63 works involve 11 distinct coauthors, and the recurring partner list — Dongyoon Han, Byeongho Heo, Sangmin Lee, Jaehui Hwang, and seven others — forms a stable cluster. All three Evolve catalog papers place Yun in a middle author position, none as corresponding author.
Activity100/100
Activity scores 100, the maximum in the snapshot. Works per year rose from 1 (2012) to 12 (2019), 13 (2021), 17 (2024) and 16 (2025), with 30 already recorded for 2026 and 63 works in the most recent three-year window — a sustained, still-accelerating cadence.
Reproducibility78/100
Reproducibility scores 78 (reproMean 78 across 2 assessed papers). In the Evolve catalog, Revisiting Complete Reasoning Traces for Post-Training is rated high reproducibility readiness and Router Prior Bias medium, giving above-average replication signals — though only two papers have been assessed so far, so coverage remains partial.
Research summary
Sangdoo Yun is a researcher at NAVER Cloud (South Korea) whose record, resolved from OpenAlex bibliographic data, shows an h-index of 31 and an i10-index of 46. Annual output has climbed steadily — from single digits before 2019 to 12 works in 2019, 13 in 2021 and 2023, 17 in 2024, and 30 already recorded for 2026 — with 63 works and 11 distinct coauthors in the most recent three-year window, earning a maximum activity score of 100. Topic assignments span five clusters, led by Advanced Neural Network Applications (71) and Domain Adaptation and Few-Shot Learning (54), while the current field split is evenly divided (0.5/0.5) between two large-language-model post-training areas: mixture-of-experts post-training and LLM post-training. The Evolve catalog holds three 2026 arXiv papers with Yun as a middle author. Router Prior Bias argues that soft-anchoring MoE routers outperforms re-applied load balancing in post-training, with careful interventional evidence though Moonlight-centric experiments. Revisiting Complete Reasoning Traces proposes trimming the middle of reasoning traces (E-SFT) across SFT, RL, and distillation, with consistent but small gains. Verification-Aware Training for Speculative Decoding awaits review. Reproducibility signals are above average (78 across two assessed papers), and the focus score of 0 reflects breadth across topic clusters rather than a narrow niche.
Collaboration pattern
Collaboration scores 50. The recent three-year window spans 63 works with 11 distinct coauthors; recurring partners include Dongyoon Han, Byeongho Heo, Jaedeok Lee, Keonwoo Kim, Yera Choi, Haanju Yoo, Geonmo Gu, HeeJae Jun, Yoohoon Kang, Sangmin Lee, and Jaehui Hwang — a stable cluster consistent with a single lab ecosystem. Across all three Evolve catalog papers Yun is a middle author, never corresponding, indicating a contributor role in the current post-training line.
Research keywords
Highlights
Router Prior Bias: soft-anchoring MoE routers in post-training
E-SFT: trimming the middle of reasoning traces
Verification-aware training for speculative decoding (2026)
A 2026 arXiv paper in the Evolve catalog with Yun as a middle author, targeting verification in speculative decoding. It has not yet been reviewed by Evolve, so no verdict, keywords, or reproducibility rating are available.
Timeline
- 2012First recorded publication appears in the OpenAlex counts-by-year series (1 work).
- 2019Output steps up to 12 works, up from 7 in 2017 and 8 in 2018.
- 202113 works, rebounding from 5 in 2020; output stays in double digits thereafter.
- 202417 works recorded — the then-highest annual count in the series.
- 202516 works; the recent three-year window reaches 63 works with 11 distinct coauthors.
- 202630 works recorded, including three arXiv papers in the Evolve catalog (two reviewed) with Yun as middle author.
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 statistics are drawn from OpenAlex only (identity resolution: openalex) and may differ from other databases.
- The output and impact radar axes are unscored because the underlying aggregates were insufficient at snapshot time.
- Only 2 catalog papers carry reproducibility assessments; Verification-Aware Training for Speculative Decoding has not yet been reviewed by Evolve.
- The record's country code (FR) conflicts with the stated institution (NAVER Cloud, South Korea); affiliation is shown as given and remains unresolved.