CAS researcher building executable benchmarks for LLM agent memory and scientific figure reconstruction
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 reliable publication-output data is available yet: OpenAlex links only one 2026 work under this name and no ORCID anchor exists. This axis would measure total papers, publication venues and first-author share once identity coverage improves.
Impact—
Impact cannot be scored yet: h-index and i10-index are 0 with essentially no accumulated citations, consistent with a record that begins in 2026. This axis would measure citation counts, h-index trajectory and uptake of the two benchmarks.
Focus0/100
The focus score is 0 because the field distribution splits evenly (0.5/0.5) between AI agents and software engineering (LLM agent memory evaluation) and machine learning (multimodal agents and evaluation), while each topic label — Software Engineering Research, Software Engineering Techniques and Practices, AI-based Problem Solving and Planning — appears only once. In practice both catalog papers share one method, executable and statistically careful agent benchmarks, so the fragmentation looks more like labeling than genuine diffusion.
Collaboration62/100
The collaboration score is 62, supported by 22 distinct coauthors across just two catalog papers — a wide network for such an early record. Recurring names include Yingcheng Shi, Yongbin Li, Chenghao Sun, Xin Chen, Xander Xu, Hu Wei, Shiwen Ni, Min Yang and Jieping Ye. Both papers are first-author but not corresponding, a pattern consistent with working inside a larger, mentor-led group.
Activity20/100
The activity score is 20. The OpenAlex-linked record lists 1 work in the last three years (countsByYear: 2026 = 1), while the Evolve catalog independently holds two 2026 arXiv papers. Either way the visible record spans a single year, so this axis reflects a young profile's publishing cadence rather than a slowdown; future cycles will show whether the two-benchmark start becomes a steady output rate.
Reproducibility55/100
The reproducibility score is 55, a mean over 2 reviewed papers, both rated 'medium' readiness by Evolve. VibeMemBench is described as a rigorous executable benchmark with explicit statistical treatment (confidence intervals) but selection-bias caveats; ReFigBench's paired study is called broadly convincing with judge-family caveats. The middling score reflects executable designs tempered by documented scope limitations, not any replication failure.
Research summary
Liyang Fan is an early-career researcher at the Chinese Academy of Sciences whose profile, resolved from OpenAlex bibliographic data, rests on two first-author arXiv benchmark papers from 2026. VibeMemBench evaluates memory systems for coding agents on real repository, SWE-bench-style tasks: Evolve's review finds a rigorous executable benchmark in which direct experience injection helps only modestly (confidence intervals cross zero) and 11 of 12 memory-system pairings fail, though selection bias limits scope. ReFigBench introduces a 1,000-figure benchmark for reconstructing scientific figures as editable PowerPoint artifacts, with a paired study showing that agent harness and workflow jointly decide quality, subject to judge-family caveats. Both papers are rated medium for reproducibility readiness (mean 55). Bibliometric coverage remains thin: one OpenAlex-linked work in 2026, zero-valued citation metrics (h-index 0, i10-index 0), and an even 0.5/0.5 split between AI-agent–software-engineering and multimodal-ML field labels. Twenty-two distinct coauthors and first-author, non-corresponding positions point to active collaboration inside a larger group. Overall this is an emerging, benchmark-focused profile whose impact metrics have not yet accumulated.
Collaboration pattern
Fan works in sizeable teams: 22 distinct coauthors appear across only two catalog papers (collaboration score 62). Recurring partners include Yingcheng Shi, Yongbin Li, Chenghao Sun, Xin Chen, Xander Xu, Hu Wei, Shiwen Ni, Min Yang, Jieping Ye, Chi Wei and Yitai Li. Both papers are first-author without corresponding-author status, suggesting contributions embedded in a mentor-led group rather than solo-led projects. The supplied coauthor list also contains the author's own name, a likely name-resolution artifact.
Research keywords
Highlights
VibeMemBench: memory systems for coding agents, mostly failing the test
ReFigBench: 1,000 scientific figures as editable PowerPoint targets
A consistent playbook: executable, statistically careful agent benchmarks
Across both catalog papers the pattern repeats: executable benchmark design, paired or comparative statistics, and candid limitation notes. This earns a reproducibility mean of 55 with medium readiness on both papers, first-author positioning on each, and a split 0.5/0.5 footprint across the AI-agent and multimodal-ML field labels.
Timeline
- 2026Publishes VibeMemBench on arXiv — a first-author benchmark evaluating memory systems for coding agents on real repository coding tasks.
- 2026Publishes ReFigBench on arXiv — a first-author 1,000-figure benchmark for scientific figure reconstruction as editable PowerPoint artifacts.
- 2026Both papers enter the Evolve catalog with medium reproducibility readiness (mean 55); the OpenAlex record shows 1 linked 2026 work and 22 distinct coauthors.
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-resolved record without an ORCID anchor; 'Liyang Fan' is a common Chinese name, so name resolution may be incomplete — the supplied coauthor list even includes the author's own name.
- Citation-based impact is not scored: h-index and i10-index are 0, reflecting a record that begins in 2026 rather than a measured absence of influence.
- OpenAlex links 1 work in 2026 while the Evolve catalog holds two 2026 arXiv papers, so coverage of this identity is likely partial.
- Qualitative assessments quote Evolve's own paper reviews; the reproducibility mean (55) covers only the 2 reviewed papers.