Beakcheol Jang

Yonsei University · South Korea

Unclaimed · AI-structured profileORCID OpenAlex

Yonsei University researcher in LLM-based and control-based time series forecasting, with high reproducibility scores.

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—

Not yet scored: the current snapshot lacks validated total-publication and per-venue aggregations for this author record. This axis would measure cumulative and annual publication volume and productivity once those inputs are complete.

Impact—

Not yet scored: per-work citation counts, field-normalized citation rates and venue-level influence are missing from the current data. This axis would measure citation-based scholarly influence and uptake of the work by other researchers once those inputs exist.

Focus0/100

Focus scores 0, the lowest value on the scale, reflecting an unusually dispersed topic profile: the five largest clusters are nearly co-equal — topic modeling (32 works), data-driven disease surveillance (19), privacy-preserving technologies (18), stock market forecasting (15) and cloud data security (15). The current field distribution is narrower, split evenly at 0.33 across three machine-learning time-series-forecasting subfields.

Collaboration46/100

Collaboration scores 46 on a 0–100 scale. The recent-three-year record lists 44 works with 9 distinct coauthors — a compact standing circle rather than a large rotating one. Recurring partners include Daeun Ji, Minkyoung Kim, Yohan Lee and Beomsoo Kim. Across the three Evolve-reviewed 2026 preprints, Jang appears as last author twice and middle author once, and as corresponding author on none.

Activity100/100

Activity scores 100, the axis maximum. Annual output climbs from 12 works (2023) and 9 (2024) to 14 (2025) and 21 (2026), the highest count in a record reaching back to 2007, and the most recent three years contribute 44 works. The profile's clearest quantitative strength is this sustained, currently peaking publication cadence.

Reproducibility85/100

Reproducibility scores 85, the mean across 3 repro-evaluated papers. Within the Evolve catalog, two of the three 2026 preprints carry high reproducibility-readiness ratings (the downside-controlled forecast combination and the coverage-aware prompt-selection paper), while CTRL is rated medium owing to missing error bars and a table mismatch. The pattern points to deliberately re-runnable experimental pipelines.

Research summary

Beakcheol Jang (Yonsei University, South Korea; ORCID 0000-0002-3911-5935) is a machine-learning researcher whose recent work centers on time series forecasting, including LLM-based and control-based approaches. OpenAlex bibliometric data shows an h-index of 26 and an i10-index of 50, with a publication record reaching back to 2007. Output cadence has accelerated markedly: per-year counts of 12 (2023), 9 (2024), 14 (2025) and 21 (2026 — the highest in the record), with 44 works in the most recent three years. The all-time topic profile is broad — topic modeling (32 works), data-driven disease surveillance (19), privacy-preserving technologies (18), stock market forecasting (15) and cloud data security (15) — while the current field distribution splits evenly across three machine-learning time-series-forecasting subfields (0.33 each). Evolve has reviewed three 2026 arXiv preprints: a downside-controlled online forecast combination that caps worst-case loss at 0.15% across 28 pairs (last author); CTRL, which casts LLM agents as label-free residual controllers for frozen forecasting backbones (middle author); and a coverage-aware prompt-selection method for LLM-based forecasting (last author). Reproducibility is a relative strength, with a mean score of 85 across three evaluated papers. The radar leaves output and impact unscored pending fuller per-work citation data, and collaboration is moderate: 9 distinct recent coauthors, with no corresponding-author role on the reviewed papers.

Collaboration pattern

The recent-three-year record shows 44 works with 9 distinct coauthors — a compact, stable circle. Recurring partners include Daeun Ji, Minkyoung Kim, Yohan Lee and Beomsoo Kim. Across Evolve's three reviewed 2026 preprints, Jang appears as last author twice and middle author once, never as corresponding author, consistent with a senior-contributor position rather than corresponding-author leadership. The collaboration radar score is 46/100.

Research keywords

Time series forecastingLLM-based forecastingForecast combinationOnline learningResidual learningTest-time adaptationDistribution shiftPrompt selectionTopic modelingPrivacy-preserving technologiesStock market forecastingData-driven disease surveillance

Highlights

Capping downside risk in online forecast combination
As last author, Jang contributed to a hedge gate placed over a frozen forecaster with two correctors that caps worst-case loss at 0.15% across 28 forecast pairs under delayed and revised outcomes. Evolve judges the construction rigorous but notes the edge over the static corrector remains unproven; reproducibility readiness is high.
CTRL: LLM agents as residual controllers for frozen forecasters
CTRL casts LLM agents as label-free residual controllers over frozen time-series-forecasting backbones, with gains that track distribution shift. Evolve flags missing error bars and a Table 3 mismatch that weaken rigor; reproducibility readiness is medium. Jang is a middle author.
Coverage-aware prompt selection beyond similarity
A parameter-free coverage regularizer for similarity-based prompt retrieval in LLM-based time series forecasting. Evolve praises the unusually honest statistical reporting but finds the aggregate gain resolves on only 1 of 24 cells; reproducibility readiness is high. Jang is last author.

Timeline

2007Earliest works appear in the OpenAlex record; annual output stays at 1–2 works through 2013.
2014First acceleration: 7 works, followed by 6 in 2015 and 6 in 2016.
2018A sustained peak begins — 14 works in 2018 and 16 in 2019, the record's pre-2026 high.
2021Output dips to 8 works, then recovers to 10 (2022) and 12 (2023).
202621 works — the record's highest annual count; includes three Evolve-reviewed arXiv preprints on LLM-based time series forecasting (last author on two, middle author on one).

Papers in the Evolve catalog

Frequent co-authors

Daeun Ji×2 Minkyoung Kim×2 Yohan Lee×2 Minkyoung Kim×1 Hyunjung Byun×1 Yohan Lee×1 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 (h-index, i10-index, per-year counts, topics, coauthor counts) come from OpenAlex only; output and impact remain unscored pending per-work citation aggregation.
Identity is resolved from public bibliographic data under ORCID 0000-0002-3911-5935; verification against the researcher's own record is advised.
Qualitative judgments on the three 2026 arXiv preprints reflect Evolve's single-pass reviews; CTRL lacks error bars and has a reported table mismatch, and one aggregate gain resolves on only 1 of 24 cells.
2026 publication counts cover a year still in progress and may rise further.