Carnegie Mellon machine learning researcher across Bayesian deep learning, LLM-based human simulation, and 3D vision
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
11
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: publication-volume data is insufficient. It would measure cumulative and annual work counts; the snapshot offers only a coarse per-year series (24 works in the last three years, 11 indexed in 2026).
Impact—
The impact axis is not yet scored: citation-coverage data is insufficient. It would measure citations, field-normalized influence, and highly cited works; the snapshot lists h-index 11 and i10 index 11 but lacks the fuller citation record needed for scoring.
Focus100/100
The focus score is 100, the maximum, reflecting a single indexed field — machine learning / Bayesian deep learning — in the field distribution. Topic clustering stays narrow-adjacent rather than scattered: 3D shape modeling and analysis (9), computer graphics and visualization (8), GAN-based image synthesis (8), face recognition and analysis (8), and advanced neural network applications (7) form five neighbouring vision/ML clusters with similar counts, so the perfect score should be read alongside this clustered breadth.
Collaboration50/100
The collaboration score is 50. The snapshot records 11 distinct coauthors, with no single dominant partner inferable from the data. The frequent-coauthor list includes José M. F. Moura, Oren Wright, Yorie Nakahira (listed twice), Fan Yang, Haoming Jing, Qiaoan Shen, Iremsu Bas, Narishige Abe, Yi Fei Cheng, and David Lindlbauer — covering both strands of the reviewed work, though the data does not map individuals to specific papers. Both catalog papers list Niinuma as a middle author, not corresponding.
Activity94/100
The activity score is 94, indicating a strongly active recent record. The counts-by-year series shows acceleration: 4 works in 2021, 2 in 2022, 4 in 2023, 8 in 2024, 5 in 2025, and 11 in 2026 — the highest year in the series. Recent-three-year output is 24 indexed works. Earlier years are sparse (single works in 2004, 2009, 2011, and 2013; two in 2010), so current output represents a marked step-up.
Reproducibility55/100
The reproducibility score is 55, computed from a single assessed paper (reproPapers 1, reproMean 55), so the estimate is thin. Evolve's review notes support a medium reading: the Bayesian smoothing paper is judged theoretically sound but with small-scale experiments and an untested distributional-target path, and the human-simulation paper relies on a five-participant, in-sample-prior dataset. Both catalog papers carry a medium repro-readiness rating, consistent with the modest score.
Research summary
Koichiro Niinuma is a machine learning researcher at Carnegie Mellon University whose recent work spans Bayesian deep learning and the evaluation of LLM-based human simulation. The Evolve catalog reviews two of his 2026 arXiv papers: one introduces full-covariance smoothing of Bayesian neural networks for online adaptation — judged theoretically sound but tested only at small scale, with the distributional-target path untested — and the other proposes a framework for measuring behavioral fidelity in long-horizon human activity simulations, built on a rare multi-day office dataset, where simulated agents match statistical priors yet fragment routines (n=5, in-sample priors). His bibliometric footprint is broader: topic clustering covers 3D shape modeling and analysis (9), computer graphics and visualization (8), GAN-based image synthesis (8), face recognition and analysis (8), and advanced neural network applications (7). The snapshot reports h-index 11, i10 index 11, 24 works in the last three years, and 11 indexed works in 2026 — the highest year recorded. Radar scores activity 94 and focus 100 (single indexed field: machine learning / Bayesian deep learning), collaboration 50 across 11 distinct coauthors, and reproducibility 55 from one assessed paper; output and impact axes lack scoring data. The profile is name-resolved from OpenAlex bibliographic data.
Collaboration pattern
Collaboration is real but moderate: the radar scores it 50, and the snapshot lists 11 distinct coauthors. The frequent-coauthor roster — José M. F. Moura, Oren Wright, Yorie Nakahira (appearing twice), Fan Yang, Haoming Jing, Qiaoan Shen, Iremsu Bas, Narishige Abe, Yi Fei Cheng, and David Lindlbauer — suggests multiple standing working groups, though the data does not tie individuals to specific papers. On both reviewed 2026 papers Niinuma is a middle author, not the corresponding author.
Research keywords
Highlights
Full-covariance smoothing brings Bayesian neural nets online
Auditing behavioral fidelity in long-horizon LLM activity simulations
Two 2026 arXiv papers bridge estimation and agent evaluation
Both Evolve-reviewed papers are dated 2026, a year with 11 indexed works — the highest in the counts-by-year series; Niinuma is a middle author on each and neither is corresponding.
Timeline
- 2004Earliest year in the counts-by-year series, with 1 indexed work; output stays sparse (1-2 per year) through 2013.
- 2019Indexed activity resumes at low volume after a 2014-2018 stretch with no indexed works, with 2 works that year.
- 20214 indexed works, opening a steadier cadence (4-2-4 across 2021-2023).
- 20248 indexed works, nearly double the 2023 count and a step toward the 2026 peak.
- 2026Highest year in the series with 11 indexed works; both Evolve-reviewed arXiv papers (Bayesian neural-network smoothing; LLM activity-simulation fidelity) are dated 2026.
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 come from an OpenAlex snapshot (identity field: openalex); h-index and i10 index are quoted as provided, with no retrieval date specified.
- The output and impact radar axes are unscored due to insufficient data; the reproducibility score rests on a single assessed paper (reproPapers 1, reproMean 55).
- Only two papers carry Evolve reviews; the other works in the counts-by-year series are not individually assessed.
- The profile is name-resolved from public bibliographic data linked to ORCID 0000-0001-8367-3988; biography beyond the listed affiliation is not independently verified.