University of Minnesota researcher advancing LLM agent memory, multi-agent debate, and game-theoretic agent coordination
4
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
26
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—
Not yet scored: the snapshot lacks a full works count and per-year productivity aggregates, so output cannot be computed. The axis would measure total publication volume and its trajectory; only the recent three-year count (39) is recorded.
Impact—
Not yet scored: citation aggregates such as total citations or field-normalized percentiles are absent from this snapshot, so influence cannot be estimated. The axis would measure scholarly influence; only the supplied h-index (26) and i10-index (60) hint at it.
Focus0/100
The focus score is 0, reflecting an exactly even spread across the snapshot's three field assignments — natural language processing (LLM agents, long-term memory), artificial intelligence (multi-agent LLM systems, algorithmic game theory), and recommender systems — each at one third. Underlying topics are similarly distributed: topic modeling (53), NLP techniques (35), information retrieval (18), network analysis (16), multimodal learning (11), indicating breadth over a single narrow track.
Collaboration67/100
Score 67, drawn from 28 distinct recent coauthors and 12 recurring collaborators, including Xuanfa Jin, Zhijian Ma, Yongcheng Zeng, Haifeng Zhang, and Meng Fang. The 39 works across the last three years pair with this broad collaborator pool, indicating steady team science; in the reviewed catalog the author sits in middle or last position, and none of the four papers lists them as corresponding author.
Activity100/100
The activity score is the maximum 100, supported by countsByYear: after double-digit annual output through most of 2009-2021, production rose to 27 works in 2022 and held at 23 in both 2023 and 2024, with 14 in 2025 and 2 already in partial 2026. The snapshot records 39 works in the last three years, a sustained recent peak.
Reproducibility55/100
Score 55, averaged over three scored papers (reproMean 55, reproPapers 3). The three LLM-agent papers carry medium reproducibility readiness — methods and benchmarks are described, but with noted gaps such as missing significance testing in R2-MAD and an externally sourced 72.2% SWE-bench comparison in Bilevel Coordinated Reflection. The mobile-game pre-training paper is rated low because data and code are closed.
Research summary
Jun Wang is a researcher at the University of Minnesota whose OpenAlex record (h-index 26, i10-index 60) shows a publishing history stretching back to 2000, with the current agenda concentrated on LLM agents. Four 2026 arXiv papers reviewed by Evolve form a coherent cluster: REALM introduces retrieval-driven memory reconsolidation for long-term agents; R2-MAD adds experience memory and confidence weighting to multi-agent debate; Bilevel Coordinated Reflection frames multi-agent LLM systems through potential games and verifier-gated reflection; and a cross-source behavior pre-training study targets mobile-game user representation. Jun Wang appears as last author on two of these papers and middle author on the other two. Underlying topic counts — topic modeling (53), natural language processing techniques (35), information retrieval and search behavior (18), complex network analysis (16), and multimodal machine learning (11) — trace a path from classic text mining toward agent-oriented NLP. Activity is the strongest axis (score 100): 39 works in the last three years, with annual output peaking at 27 works in 2022 and holding at 23 through 2023 and 2024. Because Jun Wang is a very common name, this profile is name-resolved from public bibliographic data and the listed ORCID; output and impact composite scores are not yet computed for this record.
Collaboration pattern
The record lists 28 distinct recent coauthors and 12 recurring collaborators — among them Xuanfa Jin, Zhijian Ma, Yongcheng Zeng, Haifeng Zhang, and Meng Fang — yielding a collaboration score of 67. In the reviewed catalog, Jun Wang takes last authorship on two papers and middle authorship on two, with no corresponding-author role, consistent with a senior-collaborator pattern within a rotating team of largely early-career coauthors.
Research keywords
Highlights
REALM: post-retrieval graph reconsolidation closes the memory loop for long-term LLM agents
R2-MAD: experience memory and confidence weighting sharpen multi-agent debate
Bilevel Coordinated Reflection models multi-agent LLM systems as potential games
Cross-source behavior pre-training for mobile-game user representation
Timeline
- 2026Published CM-PTM, pre-training mobile-game user representations from cross-source behavior logs (middle author).
- 2026Released Bilevel Coordinated Reflection, a game-theoretic account of multi-agent LLM systems (last author).
- 2026Released R2-MAD, adding experience memory and confidence weighting to multi-agent debate (last author).
- 2026Published REALM on retrieval-driven memory reconsolidation for long-term LLM agents (middle author).
Papers in the Evolve catalog
Retrieval-Driven Memory Reconsolidation for Long-Term LLM Agents
Remember and Reweight: Enhancing Multi-Agent Debate with Experience Memory and Confidence Estimation
Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems
User Representation via Cross Multi-source Behavior Pre-training for Mobile Games
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 a single OpenAlex record; 'Jun Wang' is a very common name, so this profile is name-resolved from public bibliographic data and the listed ORCID (0000-0002-9515-076X).
- Output and impact composite scores are not computed in this snapshot; only h-index (26) and i10-index (60) are available, and per-paper citation counts are not provided.
- All four reviewed papers are 2026 arXiv preprints; peer-review status and final venues are not reflected.
- The reproducibility score (55) averages 3 scored papers, and 2026 publication counts are partial-year.