Minkyoung Kim

Yonsei University · South Korea

Unclaimed · AI-structured profileOpenAlex

Yonsei University researcher applying LLM agents and retrieval to time series forecasting under distribution shift.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) 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 data is insufficient for scoring: h-index and i10-index are recorded as 0 and only one 2026 work is indexed in OpenAlex, so no publication-volume score can be computed yet. This axis would measure cumulative publication count and productivity across years.

Impact—

Impact data is insufficient for scoring: the profile shows h-index 0 and i10-index 0 with no citation time series available, so scholarly influence cannot be estimated. This axis would measure citations, uptake, and the reach of the published work.

Focus0/100

Focus scores 0, the lowest band, but the underlying distribution is narrow rather than scattered: fieldDist splits evenly at 0.5 for LLM-based time series forecasting and 0.5 for time series forecasting, and three topics (topic modeling, forecasting techniques, time series analysis) each carry a count of 1. The zero reflects an agenda still crystallizing around one problem area, not genuinely unfocused breadth.

Collaboration39/100

Collaboration scores 39 on the 0-100 axis, with 6 distinct co-authors recorded in the raw profile. The pattern is small-team: Daeun Ji, Yohan Lee, Beakcheol Jang, Dongkuk Kim, and Beomsoo Kim recur across the reviewed catalog, and Kim moves between first authorship on CTRL and a middle position on the coverage-aware paper, consistent with work inside a stable Yonsei-centered group rather than broad external networking.

Activity20/100

Activity scores 20, and the raw counts explain why: recent3yWorks is 1 and countsByYear lists a single work in 2026. The visible record is a narrow, recent window — two arXiv preprints dated 2026 in Evolve's reviewed catalog — rather than a multi-year output stream. As an early-stage profile, this axis should be re-read once additional indexed years accumulate.

Reproducibility78/100

Reproducibility is the strongest axis at 78, computed as the mean over 2 reviewed papers. Evolve's readiness labels support this: the coverage-aware prompt-selection paper is rated high and CTRL medium. The same reviews flag statistical weaknesses — missing error bars and a Table 3 mismatch on CTRL, and an aggregate gain resolving on only 1 of 24 cells on the other paper — so the score reflects artifact readiness more than statistical airtightness.

Research summary

Minkyoung Kim is a machine learning researcher at Yonsei University, South Korea, whose visible record centers on making large language models useful for time series forecasting. Two 2026 arXiv preprints define the profile. The first, CTRL, with Kim as first author, casts LLM agents as label-free residual controllers over frozen forecasting backbones, combining residual learning, STL decomposition, and test-time adaptation to handle distribution shift; Evolve's review finds the gains track shift severity but notes missing error bars and a Table 3 mismatch. The second, Beyond Similarity, adds a parameter-free coverage regularizer to counter prompt selection bias in similarity-based retrieval for LLM-based forecasting; it is praised for unusually honest statistics even though its aggregate gain resolves on only 1 of 24 cells. Bibliometric coverage is thin and recent: OpenAlex indexes 1 work in 2026, with h-index and i10-index both recorded as 0, so output and impact are left unscored. The record is tightly focused — an even 0.5/0.5 split between LLM-based and general time series forecasting — and reproducibility is the strongest measured axis at 78, the mean over 2 reviewed papers rated medium and high for readiness. Six distinct co-authors, including Daeun Ji, Yohan Lee, and Beakcheol Jang, point to a small stable team rather than broad external networking. This is an early-stage profile; conclusions should be revisited as more indexed years accrue.

Collaboration pattern

The record shows a compact, repeat collaboration pattern: 6 distinct co-authors, with Daeun Ji, Yohan Lee, Beakcheol Jang, Dongkuk Kim, and Beomsoo Kim recurring across the two reviewed 2026 papers. Kim held first authorship on CTRL and a middle position on Beyond Similarity, neither as corresponding author, consistent with contributing inside a stable Yonsei-centered team rather than an externally networked one.

Research keywords

LLM agentsTime series forecastingResidual learningTest-time adaptationDistribution shiftSTL decompositionPrompt selection biasCoverage regularizerSimilarity-based retrievalDiversity-aware retrievalSemantic anchorsTopic modeling

Highlights

LLM agents as label-free residual controllers for frozen forecasting backbones
First-author 2026 arXiv preprint CTRL frames LLM agents as label-free residual controllers over frozen time series forecasting backbones, combining residual learning, STL decomposition, and test-time adaptation to track distribution shift. Evolve's review credits the shift-tracking gains but flags missing error bars and a Table 3 mismatch; reproducibility readiness is rated medium.
Coverage-aware prompt selection against similarity bias in LLM forecasting
Co-authored 2026 preprint Beyond Similarity introduces a parameter-free coverage regularizer for LLM prompt retrieval in time series forecasting, countering prompt selection bias in similarity-based retrieval with semantic anchors and diversity-aware retrieval. Evolve notes unusually honest statistics, though the aggregate gain resolves on only 1 of 24 cells; reproducibility readiness is high.
A young record concentrated on rigor-checkable LLM-for-forecasting methods
Both reviewed 2026 preprints sit in LLM-based time series forecasting and both were audited by Evolve, yielding a reproducibility mean of 78 over 2 papers (readiness medium and high). The shared thread is methods that expose themselves to verification, while the same reviews surface missing error bars and fragile aggregate gains as open rigor gaps.

Timeline

2026Posts first-author arXiv preprint CTRL, casting LLM agents as label-free residual controllers for time series forecasting with residual learning and STL decomposition.
2026Co-authors Beyond Similarity, a coverage-aware prompt-selection method for LLM-based forecasting, in a middle author position.
2026OpenAlex logs the profile's first indexed work (1 work in 2026), and both catalog papers enter Evolve review with a reproducibility mean of 78.

Papers in the Evolve catalog

2026-09-20first author
CTRL: Control-Based Time Series Forecasting with LLM-Guided Residual Learning
2026-09-19
Beyond Similarity: Coverage-Aware Prompt Selection for Time Series Forecasting with LLMs

Frequent co-authors

Daeun Ji×2 Yohan Lee×2 Beakcheol Jang×2 Dongkuk Kim×1 Beomsoo Kim×1 Beomsoo Kim×1

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 indicators are computed from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may merge or split namesakes.
h-index and i10-index are recorded as 0 and no citation time series is available, so output and impact are left unscored rather than estimated.
The evidence base is two 2026 arXiv preprints reviewed by Evolve; peer-reviewed versions and earlier work are not captured.
Only two reviewed papers back the highlights and dimension narratives, so all conclusions should be read as early-stage and re-checked as data accrues.