Yumin Choi
Korea Advanced Institute of Science and Technology · South Korea
Yumin Choi — KAIST researcher on self-evolving LLM agents for trading and code, with a highly focused early record.
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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—
This axis would measure publication volume and authorship role (first/corresponding authorships). No output score is computed yet: the OpenAlex snapshot indexes only 8 works, all dated 2026, with h-index 0 and i10-index 0 — too sparse a record to score reliably.
Impact—
Impact would capture citation-based reach (citations, h-index, i10-index). The current record reports h-index 0 and i10-index 0 across 8 indexed 2026 works, so no meaningful impact score exists yet; the record is simply too new to interpret.
Focus100/100
The focus dimension scores 100, the maximum, indicating an exceptionally concentrated record. The field distribution lists a single field — machine learning for LLM agents in financial trading — and the topic profile is led by Topic Modeling (5 works) with smaller clusters in software engineering research, reinforcement learning in robotics, AI ethics, and AI-based planning (2 each), outlining a coherent agent-and-learning core.
Collaboration41/100
Collaboration scores 41 on a 0-100 scale, a mid-range level supported by 7 distinct coauthors in the indexed record. The recurring partner list is short — Sehee Kim, Minki Kang, Sung Ju Hwang, Daegyu Sung, Yukyeong Lee, and Geon Park — consistent with a small, stable team pattern around the KAIST group rather than a wide external network.
Activity64/100
Activity scores 64, a solid mid-high level. All 8 indexed works fall in 2026 (recent-3-year works: 8), meaning the entire visible output is concentrated in a single burst this year. That recency makes the pace look strong, but the short window means the activity trend cannot yet be read across multiple years.
Reproducibility55/100
Reproducibility scores 55, computed from a single reviewed paper (reproPapers: 1, mean readiness 55/100). That paper — EvolveTrade — carries a medium repro-readiness rating; Evolve's verdict notes 3 runs, 20-day evaluation windows, and no significance tests. A one-paper basis means this score should be read as indicative, not settled.
Research summary
Yumin Choi is a KAIST-affiliated researcher whose profile is name-resolved from OpenAlex public bibliographic data. The indexed record is young and dense: all 8 works are dated 2026, and citation indices (h-index 0, i10-index 0) have not yet accumulated, so output and impact scores are not computed. The research core is LLM-based agents. In the Evolve catalog, Choi appears as a middle author on EvolveTrade (2026, arXiv), which proposes experience-driven policy refinement that lets self-evolving LLM trading agents rewrite their own system prompts for portfolio management in non-stationary markets; Evolve's verdict finds the gains suggestive — best in 3 of 6 regimes — but thin, with 3 runs, 20-day windows, and no significance testing (medium reproducibility readiness). A second 2026 arXiv paper, Super Library Agent, extends agent systems toward joint generation and maintenance of multiple applications beyond a single codebase. Bibliometrically, the record is strikingly concentrated: focus scores 100/100, with a single listed field (machine learning for LLM agents in financial trading) and Topic Modeling leading the topic counts (5), followed by software engineering research, reinforcement learning in robotics, AI ethics, and AI-based planning (2 each). Activity scores 64, collaboration 41 across 7 distinct coauthors — including Sehee Kim, Minki Kang, Sung Ju Hwang, Daegyu Sung, Yukyeong Lee, and Geon Park — and reproducibility 55 from one reviewed paper. This is an early-stage profile likely to change quickly.
Collaboration pattern
The indexed record lists 7 distinct coauthors, and the collaboration score is 41 — a modest, small-team pattern. Recurring names include Sehee Kim, Minki Kang, Sung Ju Hwang, Daegyu Sung, Yukyeong Lee, and Geon Park. In both cataloged papers Choi sits in a middle-author, non-corresponding position, consistent with contributing specialist roles inside larger KAIST-led teams rather than leading an independent group so far.
Research keywords
Highlights
Self-rewriting prompts for LLM trading agents
Suggestive gains with thin evidence
Super Library Agent: beyond a single codebase
A second 2026 arXiv preprint, with Choi as a middle author, proposes joint generation and maintenance of multiple applications by one agent system; it is listed in the catalog without an Evolve verdict yet.
Timeline
- 2026Record burst: OpenAlex indexes 8 works for 2026, the only year with counts — the visible record is a single-year burst.
- 2026EvolveTrade preprint: joins a 2026 arXiv paper on experience-driven policy refinement for self-evolving LLM trading agents as a middle author.
- 2026Super Library Agent preprint: second 2026 arXiv paper, on joint generation and maintenance of multiple applications beyond a single codebase.
- 2026Early metrics form: focus 100, activity 64, collaboration 41, reproducibility 55 computed; h-index and i10-index stand at 0.
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 are computed from an OpenAlex identity snapshot only; no ORCID is linked, so name ambiguity is possible.
- Citation measures (h-index 0, i10-index 0) and the output/impact dimensions reflect a record whose 8 indexed works are all dated 2026 — too recent to interpret.
- The reproducibility score (55) rests on a single reviewed paper and should be read as indicative only.
- One cataloged paper (Super Library Agent) has not been reviewed by Evolve and carries no verdict, keywords, or repro-readiness rating.