Weinan Zhang · Shanghai Jiao Tong University — LLM agents, long-term memory, and multi-agent AI safety
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
60
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—
No output score was computed in this snapshot, so publication volume is unscored. This axis would measure total output volume, venue spread, and productivity trend; only partial signals — e.g., 146 works in the last three years — are on record.
Impact—
No impact score was computed in this snapshot, so citation-based influence is unscored. The axis would measure citation volume and scholarly uptake; the raw record lists an h-index of 60 and an i10-index of 195, but no composite impact score.
Focus0/100
The focus score is 0, the lowest value, reflecting a highly dispersed topical footprint: the raw topic profile spans five sizable clusters — Recommender Systems (169), Information Retrieval (138), Reinforcement Learning in Robotics (99), Advanced Bandit Algorithms (98), and Topic Modeling (86) — while reviewed 2026 output splits evenly (0.5/0.5) between NLP (LLM agents, long-term memory) and computational social science / multi-agent AI safety.
Collaboration61/100
The collaboration score is 61, indicating a solid but not maximal collaborative footprint. The raw record lists 20 distinct coauthors in the recent three-year window and a recurring pool of 12 frequent collaborators, including Jun Wang, Weiwen Liu, Jianghao Lin, Muning Wen, and Shangding Gu. Author positions in the reviewed catalog also vary — last author on two 2026 papers and middle author on one — consistent with a senior supervisory role alongside active peer collaborations.
Activity100/100
Activity scores a perfect 100. The per-year work counts climb almost monotonically from 17 works in 2016 and 41 in 2018 to 55 in 2023, 61 in 2024, and 67 in 2025, with 18 works already recorded in the partial 2026 year; the record also shows 146 works in the last three years. This is an unusually steep, still-rising productivity curve.
Reproducibility55/100
Reproducibility scores 55, computed from 2 reviewed papers with a mean readiness of 55. Both carry a medium rating: REALM's reconsolidation pipeline and WolfSociety's agent-society simulations are judged careful but partial — Evolve notes the reconsolidation step adds only about 2 points and the collapse exponent is cutoff-sensitive — so replication likely requires nontrivial additional effort.
Research summary
Weinan Zhang is a researcher at Shanghai Jiao Tong University whose cataloged 2026 work concentrates on large language model agents — long-term memory, retrieval-augmented generation, and process supervision — alongside computational social science and multi-agent AI safety. The OpenAlex-derived record is exceptionally active: an h-index of 60, an i10-index of 195, 146 works in the last three years, and annual output climbing from 41 works in 2018 to 67 in 2025, with 18 already logged in partial 2026 — a perfect activity score of 100. Topical reach is broad: the largest clusters are Recommender Systems (169), Information Retrieval (138), Reinforcement Learning in Robotics (99), Advanced Bandit Algorithms (98), and Topic Modeling (86), producing a focus score of 0 and an even 0.5/0.5 split of reviewed 2026 output between NLP (LLM agents, long-term memory) and computational social science / multi-agent AI safety. Collaboration scores 61, with 20 distinct recent coauthors and a recurring group including Jun Wang, Weiwen Liu, and Jianghao Lin. Two 2026 arXiv papers carry Evolve verdicts: REALM's retrieval-driven memory reconsolidation closes the memory loop for long-term agents with strong benchmarks, while WolfSociety locates a collapse boundary of αc ∝ N^−0.22 for harmful-agent scaling in financial agent societies. Both are rated medium for reproducibility (mean 55).
Collaboration pattern
The collaboration score of 61 rests on 20 distinct coauthors in the recent three-year window and a stable recurring pool of 12, including Jun Wang, Weiwen Liu, Jianghao Lin, Muning Wen, Shangding Gu, Xinbei Ma, and Xin Jiang. Position patterns in the 2026 catalog — last author on two papers, middle author on one — suggest a senior-supervisor role on LLM-agent projects alongside peer-level collaboration on simulation-based safety studies.
Research keywords
Highlights
REALM: closing the memory loop for long-term LLM agents
WolfSociety: a scaling law for collapse in financial agent societies
From final artifacts to trajectories: retrospective process supervision
Zhang's other last-author 2026 arXiv paper shifts supervision from final outputs to whole reasoning trajectories for evidence-grounded long-form generation. Evolve has not yet published a verdict, keywords, or reproducibility rating for this paper.
Timeline
- 2026Last-author arXiv release of 'Retrieval-Driven Memory Reconsolidation for Long-Term LLM Agents', later reviewed by Evolve with a medium reproducibility rating.
- 2026Middle-author on 'WolfSociety: Understanding Collective Risk from Harmful-Agent Scaling in Financial Agent Societies', an Evolve-reviewed agent-society simulation study.
- 2026Last-author on 'From Final Artifacts to Trajectories: Retrospective Process Supervision for Evidence-Grounded Long-Form Generation'; no Evolve verdict on record yet.
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 are derived from OpenAlex records only; citation and topic coverage may be incomplete.
- Output and impact dimension scores were not computed in this snapshot; the h-index (60) and i10-index (195) are quoted verbatim from the raw record.
- Qualitative review evidence covers only three 2026 arXiv papers from the Evolve catalog; one has no verdict, keywords, or reproducibility rating yet, and the 2026 work count (18) is a partial-year figure.
- 'Weinan Zhang' is a common name; this profile is name-resolved from public bibliographic data (ORCID 0000-0002-0127-2425).