Researcher at Renmin University of China working on memory-efficient long-horizon LLM agents and context compression.
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 OpenAlex snapshot shows h-index 0, i10-index 0, and only three recent works. This axis would measure cumulative publication volume and productivity trajectory; the profile is too new for volume-based scoring to mean much.
Impact—
Not yet scored: citation-based impact cannot be computed — h-index and i10-index are 0 in the snapshot and the reviewed papers are fresh 2026 preprints. This axis would measure citations, influence on follow-up work, and field recognition.
Focus0/100
The focus score reads 0/100, but the raw field distribution tells a narrower story: the three recent works split evenly (0.333 each) across labels that are effectively one area — machine learning for LLM agents and efficient context management with NLP overlap. Topic Modeling appears three times among assigned topics. With only three indexed works, the 0 score reflects thin data; the visible agenda is tightly single-themed.
Collaboration53/100
Collaboration scores 53/100. OpenAlex records 13 distinct coauthors across only three recent works — a fairly wide network for the output volume. All three cataloged papers place Ma in a middle author position with no corresponding role, consistent with contributing inside a larger team; recurring coauthors include Mingxuan Wang, Hongyue Chen, Yinglong Guo, Chao Ning, and Guorun Yao.
Activity40/100
Activity scores 40/100. The OpenAlex counts-by-year record shows a single indexed year, 2026, with three works, and recent3yWorks is 3; no earlier output appears in the snapshot. All three items are arXiv preprints from the same cycle, so the modest score reflects a short indexed history rather than dormancy.
Reproducibility43/100
Reproducibility scores 43/100, the reproMean across three scored papers. Evolve rates StateComp and the Memory Control Signals (PaMER) paper as medium readiness and DRSR as low. Recurring concerns are single-run evaluations without error bars and labels produced by LLMs, so the headline gains (52.27%, ~73%, -20.82% token cuts) await multi-seed confirmation.
Research summary
Yanbiao Ma is a machine-learning researcher at Renmin University of China whose recent work centers on memory and context management for long-horizon LLM agents. The Evolve-reviewed catalog links three 2026 arXiv preprints, in each of which Ma is a middle author: StateComp, which learns when to compress agent history and reports a 52.27% token reduction at near-flat reward; Memory Control Signals Emerge Before Action, which shows hidden states encode memory needs before action (AUROC .831/.765) and uses PaMER to cut tokens ~73% at near-par scores; and DRSR, which learns set-level deletion risk via counterfactual supervision, reporting +0.103 reward at -20.82% tokens. Together they form a tightly single-themed agenda: deciding what history an agent should keep, when to compress it, and what deletion risk means. OpenAlex bibliometrics are early-stage — h-index and i10-index are 0, three works in the last three years, 13 distinct coauthors, and topic assignments spanning topic modeling, multimodal ML, XAI, and materials-science ML. Evolve rates reproducibility readiness medium for two papers and low for DRSR, citing single-run evaluations and absent error bars; output and impact dimensions remain uncomputed.
Collaboration pattern
Across the three cataloged 2026 papers Ma sits in a middle author position and is not corresponding author, consistent with a contributor role inside a team. OpenAlex records 13 distinct coauthors over the last three years. Recurring names include Mingxuan Wang, Hongyue Chen, Yinglong Guo, Chao Ning, and Guorun Yao, with Jungong Han, Fei Luo, and Bo Wang also appearing. The frequent-coauthor list even contains Ma's own name, suggesting some entity-resolution noise.
Research keywords
Highlights
StateComp: learning when to compress agent history
PaMER: memory-control signals emerge before action
DRSR: set-level deletion risk for long-horizon agents
Timeline
- 2026StateComp preprint appears on arXiv, learning state-based compression timing that cuts agent tokens 52.27% at near-flat reward.
- 2026Memory Control Signals (PaMER) preprint reports pre-action encoding of memory needs in hidden states (AUROC .831/.765) and ~73% token cuts.
- 2026DRSR preprint introduces counterfactually supervised set-level deletion risk, reporting +0.103 reward at -20.82% tokens.
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.
- All bibliometrics come from OpenAlex only; no ORCID is linked, so the profile is name-resolved from public bibliographic data and may absorb or omit namesakes' work.
- Output and impact dimensions are missing because h-index and i10-index are 0 in the snapshot — an early-career signal, not a measured zero of merit.
- Every reviewed paper is a 2026 arXiv preprint; no peer-reviewed venue or citation record is captured yet.
- The frequent-coauthor list includes Yanbiao Ma's own name, indicating entity-resolution noise in the underlying coauthor data.