KAIST researcher bridging multimodal video memory agents and LLM benchmarks over real-world wearable health data
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
10
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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: consolidated publication-volume data for this identity is insufficient. This axis would measure annual and cumulative productivity — total works, per-year counts and first-author share — once the OpenAlex record is fully disambiguated.
Impact—
Not yet scored: citation-based impact data is not available in this snapshot. The axis would measure citations per paper, field-normalized citation percentiles and the uptake of this work in follow-up research.
Focus0/100
Scores 0, the lowest topical concentration: OpenAlex topics spread across five clusters — Multimodal Machine Learning Applications and CCD/CMOS Imaging Sensors (6 each), Rock Mechanics and Modeling (5), Information Retrieval and Search Behavior (3), Fatigue and Fracture Mechanics (3) — and the field split is even at 0.5/0.5 between computer vision and machine learning. The two reviewed 2026 papers are themselves coherent; the dispersion likely reflects a broad or namesake-mixed record.
Collaboration56/100
Scores 56: 16 distinct coauthors across 13 recent-three-year works, a mid-range collaborative breadth. The recurring roster includes Dohwan Ko and Hyunwoo J. Kim alongside Pierce Chuang, Ashish Shenoy, Debojeet Chatterjee, Yichao Lu, Seungwhan Moon, Xin Luna Dong, Xilun Chen and Vikas Bhardwaj. Authorship position flexes between the two reviewed papers — first author on WearableQA, middle author on SVMemAgent — mixing leading roles with contributions to group-built systems.
Activity77/100
Scores 77: output is current and steady — 13 works in the recent three-year window, with counts-by-year of 5 in 2024, 5 in 2025 and 3 already recorded in 2026, including the two 2026 arXiv preprints in the Evolve catalog. Counts-by-year also list sparse earlier years (1993-2007), but the recent run shows a consistently active pace.
Reproducibility78/100
Scores 78, a mean across 2 screened papers. WearableQA rates high readiness as a dual-grounded benchmark with a releasable design, while SVMemAgent rates medium, reflecting its reinforcement-learning training pipeline (Group Relative Policy Optimization). Evolve flags for WearableQA: gold-label noise and QC-model overlap.
Research summary
Ji Soo Lee (Korea Advanced Institute of Science and Technology, South Korea; ORCID 0009-0003-9055-5236) is an early-career researcher working where multimodal machine learning meets LLM evaluation for digital health. The Evolve-reviewed record centers on two 2026 arXiv preprints: WearableQA, a first-author benchmark of 4,084 multiple-choice questions built over real wearable data from 200 users, which separates 14 LLMs across a 19.6-72.9% accuracy range; and SVMemAgent, a co-authored, GRPO-trained streaming video memory agent for query-agnostic online frame selection. OpenAlex bibliometrics — name-resolved from public bibliographic data — list an h-index of 10 and i10-index of 10, 13 works in the recent three-year window, and 16 distinct coauthors, with the field split even at 0.5 computer vision / 0.5 machine learning. Activity is current and steady: counts-by-year show 5 works in each of 2024 and 2025 and 3 already in 2026. Evolve's reproducibility screening across the two catalog papers averages 78, with WearableQA rated high readiness and SVMemAgent medium. Output and impact dimensions are not yet scored, and the sparse 1993-2007 portion of the count history (imaging sensors, rock mechanics, fatigue topics) likely reflects namesake records rather than this researcher's own career.
Collaboration pattern
Sixteen distinct coauthors appear across 13 recent-three-year works (collaboration score 56). The recurring roster includes Dohwan Ko and Hyunwoo J. Kim alongside Pierce Chuang, Ashish Shenoy, Debojeet Chatterjee, Yichao Lu, Seungwhan Moon, Xin Luna Dong, Xilun Chen and Vikas Bhardwaj. Authorship position flexes — first author on WearableQA, middle author on SVMemAgent — so the pattern mixes leading small-team benchmark efforts with contributing to larger group-built systems.
Research keywords
Highlights
First-author WearableQA: health reasoning benchmark over real wearable data
GRPO-trained streaming video memory agent for query-agnostic frame selection
A dual-track 2026 output: benchmark building and RL-trained video agents
The two reviewed papers pair careful benchmark construction (WearableQA, high reproducibility readiness) with reinforcement-learning systems work (SVMemAgent, GRPO), matching the even 0.5/0.5 split between computer vision and machine learning in the OpenAlex field distribution.
Timeline
- 2026First-author WearableQA benchmark posted to arXiv: 4,084 MCQs over 200 users' wearable data, separating 14 LLMs across 19.6-72.9%.
- 2026Co-authored SVMemAgent preprint released on arXiv: a GRPO-trained streaming video memory agent for query-agnostic online frame selection.
- 2026OpenAlex counts-by-year record 3 works for 2026 to date, extending the 5-work years of 2024 and 2025 and bringing the recent three-year total to 13 works.
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 from OpenAlex only; the profile is name-resolved from public bibliographic data for a common Korean name, with ORCID 0009-0003-9055-5236 on file.
- Sparse count-by-year records from 1993-2007 under imaging-sensor, rock-mechanics and fatigue topics likely reflect namesake works merged into the same OpenAlex identity; no early-career facts are asserted.
- Output and impact dimensions are unscored; citation aggregates beyond h-index 10 and i10-index 10 were not available in this snapshot.
- Both catalog papers are 2026 arXiv preprints; peer-review status, venue acceptance and funding information are not captured.