Makoto Fukushima
Honda Research Institute Japan Co., Ltd. (Japan) · Japan
Honda Research Institute researcher quantifying collective truth-finding and implicit communication in LLM networks
2
papers here
2
reviewed by Evolve
2
first-author
1
corresponding
—
works · OpenAlex
0
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—
Measures publication volume and cumulative productivity. It cannot be scored yet: the snapshot records an h-index and i10-index of 0 with only 5 recent works, so output coverage in the data is insufficient for a meaningful score.
Impact—
Measures citation-based influence such as citation counts, h-index trajectories, and uptake of methods by others. No citation-impact data is available in this snapshot (h-index 0, i10-index 0), so the axis is left unscored rather than estimated.
Focus0/100
OpenAlex splits this author's field distribution exactly evenly — 0.5 to cooperative multi-agent systems and human-AI evaluation and 0.5 to computational social science (LLM multi-agent systems / statistical mechanics) — and the recorded topic list spans five areas, led by complex network analysis techniques (4) and neural dynamics and brain function (3). The focus score of 0 reflects this dual-track, breadth-heavy profile rather than a single dominant line of work.
Collaboration22/100
The raw OpenAlex record lists only 2 distinct coauthors, with Hua-Dong Xiong and Ehsan Moradi Pari as frequent collaborators, yielding a low collaboration score of 22. Both cataloged papers are first-authored (one also corresponding), consistent with a small, tightly-knit working pattern rather than large distributed teams. The sample is small, so this reading may shift as more works are indexed.
Activity52/100
OpenAlex records 5 works in the recent three-year window, all in 2026 (countsByYear: 2026 — 5), giving a mid-range activity score of 52. The two Evolve-reviewed 2026 arXiv preprints fall inside this window. Because indexed history before 2026 is absent, activity is measured over a very short observable span and should be read cautiously.
Reproducibility78/100
Evolve's reproducibility assessment averages 78 across the 2 cataloged papers. The Convention Gap paper is rated high readiness, offering an exactly computable measure validated by an off-belief learning dose-response experiment; the message-capacity paper is rated medium, with blind capacity prediction failing but honest failure analysis and qwen3:8b validation (15/16). This is a strong, evidence-first reproducibility signal on a small base.
Research summary
Makoto Fukushima is a researcher at Honda Research Institute Japan working at the intersection of cooperative multi-agent AI and computational social science. An OpenAlex-resolved profile shows an even field split (0.5/0.5) between cooperative multi-agent systems and human-AI evaluation and computational social science (LLM multi-agent systems / statistical mechanics), with recorded topic concentrations in complex network analysis techniques (4) and neural dynamics and brain function (3). Two 2026 arXiv preprints, both first-authored, are in the Evolve catalog. 'The Convention Gap' proposes an exactly computable measure of implicit communication in Hanabi, validated through an off-belief learning dose-response experiment, and is rated high for reproducibility readiness. 'Message capacity and claim wording set the transition points of collective truth-finding in language-model networks' models LLM collectives as binary-neuron networks; its blind capacity prediction failed, but the honest failure analysis and qwen3:8b validation (15/16) were rated favorably, at medium reproducibility readiness. Bibliometrics remain thin: OpenAlex records 5 works in the recent three-year window (all dated 2026), 2 distinct coauthors (Hua-Dong Xiong and Ehsan Moradi Pari recur), and an h-index and i10-index of 0, so output and impact axes are flagged as lacking sufficient data. The reproducibility axis scores 78 (mean over 2 assessed papers), consistent with a measurable-methods, evidence-first style.
Collaboration pattern
Collaboration is narrow: OpenAlex lists just 2 distinct coauthors, and Hua-Dong Xiong and Ehsan Moradi Pari recur as frequent collaborators. Both cataloged 2026 papers are first-authored — one also corresponding — suggesting a small, hands-on team style at Honda Research Institute Japan in which the author drives analysis and writing. The low collaboration score (22) matches this compact pattern and should be revisited as more indexed output appears.
Research keywords
Highlights
An exactly computable measure of implicit communication in Hanabi
Binary-neuron network models of collective truth-finding in LLM societies
A measurable-methods, evidence-first working style
Across the 2 assessed papers, Evolve's reproducibility mean is 78, with one high and one medium readiness rating — both first-authored works that define an exact metric or honestly report a failed prediction, suggesting a preference for quantifiable, testable claims.
Timeline
- 2026First-authored 'The Convention Gap: Towards Measuring Implicit Communication in Cooperative AI Evaluation' posted to arXiv, introducing an exactly computable convention-gap metric for Hanabi, validated by an off-belief learning dose-response experiment.
- 2026First-authored and corresponding 'Message capacity and claim wording set the transition points of collective truth-finding in language-model networks' posted to arXiv, modeling LLM collectives as binary-neuron networks with qwen3:8b validation (15/16).
- 2026OpenAlex snapshot records 5 works in the recent three-year window, all dated 2026, with 2 distinct coauthors and an h-index of 0 — a short but active indexed record.
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.
- Bibliometric metrics (topics, counts, coauthors, indices) come from OpenAlex only; output and impact axes lack usable citation data (h-index and i10-index recorded as 0).
- 'Makoto Fukushima' is name-resolved from public bibliographic data; the OpenAlex topic profile (complex networks, neural dynamics) differs from the reviewed 2026 LLM multi-agent papers, so records may blend homonymous authors.
- Only 2 Evolve-reviewed papers underpin the qualitative assessment; no ORCID, career, award, or grant information is available in the input data.
- The indexed record spans only 2026 (5 works), so activity, focus, and collaboration scores rest on a very small observable base.