Researcher at Fudan University focused on memory for LLM agents: parametric memory and memory-calibration benchmarks
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 data is available in this snapshot: publication counts, venue history and productivity trends were not provided. This axis would measure total publication volume and its distribution across venues and years.
Impact—
Impact cannot be scored yet: no citation counts, h-index or downstream-usage signals exist in the input data. This axis would measure how widely the work is cited and built upon by other researchers.
Focus0/100
Evolve's radar records a focus score of 0, derived from a field distribution that splits evenly at 0.5/0.5 between 'machine learning / natural language processing (LLM agent memory)' and 'machine learning (LLM agents / NLP)'. In practice, both reviewed papers converge on memory for LLM agents — recurrent parametric memory in RPMem and memory-use calibration in MemCalib — so the low score reflects the even two-label split rather than visibly scattered topics.
Collaboration58/100
Collaboration scores 58 on Evolve's 0-100 radar, with 17 distinct coauthors recorded across the reviewed record. The recurring partner group includes Liang Dong, Fugen Yao, Jian Xu, Guanjun Jiang, Han Zhang, Yifei Zhao, Ruike Cao, Yinsheng Li and Li Xiao. The pattern suggests steady team-based work — first-author leadership on RPMem alongside a middle-author contribution to MemCalib — rather than large-scale multi-group networks.
Activity—
Activity data is missing: no per-year publication record beyond the two 2026 arXiv preprints is available. This axis would measure recent publication cadence and sustained research engagement over time.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers assessed, both rated medium readiness. Evolve's reviews note thorough ablations on both papers but flag caveats: RPMem's gains are strong on PERMA yet modest elsewhere with multiple-choice-only evaluation, and MemCalib's LLM-judge loop introduces circularity. These documented limitations inform the mid-range score; released code or data was not confirmed in this snapshot.
Research summary
Fanyu Zhao is a researcher at Fudan University whose recent work centers on memory for large language model (LLM) agents. The Evolve catalog includes two 2026 arXiv preprints. The first-authored RPMem proposes a long-term recurrent parametric memory consolidated across sessions, drawing on hypernetworks and LoRA; Evolve's review credits strong, well-ablated gains on PERMA but notes modest results on other benchmarks and evaluation restricted to multiple-choice formats. The second paper, MemCalib, a middle-author contribution, benchmarks and optimizes memory use in LLM agents on a 15,000-example suite with counterfactual-credit reinforcement learning; Evolve calls it rigorous and thoroughly ablated, while cautioning that an LLM-judge loop weakens the reliability of its conclusions. Bibliometrically, Zhao's reviewed output splits evenly (0.5/0.5) between 'machine learning / natural language processing (LLM agent memory)' and 'machine learning (LLM agents / NLP)', which maps to a focus score of 0 in the radar. Collaboration is moderate at 58, with 17 distinct coauthors recorded, including recurring partners such as Liang Dong, Fugen Yao, Jian Xu, Guanjun Jiang and Han Zhang. Both reviewed papers carry medium reproducibility readiness, averaging 55. No citation, publication-volume or per-year activity data is available in this snapshot, so output and impact cannot yet be characterized.
Collaboration pattern
Evolve records 17 distinct coauthors for Fanyu Zhao and a moderate collaboration score of 58. The recurring group includes Liang Dong, Fugen Yao, Jian Xu, Guanjun Jiang, Han Zhang, Yifei Zhao, Ruike Cao, Yinsheng Li and Li Xiao. The catalog pattern — first-author leadership on RPMem plus a middle-author role on MemCalib, with overlapping keyword sets — points to a stable team working within a single LLM-agent-memory research line rather than dispersed multi-group ties.
Research keywords
Highlights
First-authored recurrent parametric memory for LLM agents
Memory-use calibration benchmark with counterfactual-credit RL
Timeline
- 2026First-authored RPMem appears on arXiv, proposing recurrent parametric memory consolidated across sessions for LLM agents.
- 2026Co-authors MemCalib (middle author), a 15,000-example benchmark with counterfactual-credit RL for memory calibration.
- 2026Both preprints enter the Evolve catalog with medium reproducibility readiness, averaging 55, concentrated in LLM agent memory.
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 dimensions are computed from a narrow snapshot; only the two Evolve-reviewed arXiv preprints are covered, and other publications are not reflected.
- Output, impact and activity axes have no underlying data (no citation counts, publication volume or per-year records), so they are left unscored.
- Profile is name-resolved from public bibliographic data; no ORCID was available and 'Fanyu Zhao' is a common name, so identity consolidation is not guaranteed.
- Reproducibility score (mean 55) averages only 2 assessed papers, both rated medium readiness; code and data availability were not confirmed.