Guorun Yao
—
Researcher on memory compression and control for long-horizon LLM agents; four 2026 arXiv preprints, middle authorship
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-01) 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 score yet: the record holds only four 2026 arXiv preprints with no aggregated publication counts. This axis would measure research output — paper volume, venue types, and productivity over time — once a fuller publication record is linked.
Impact—
No score yet: no citation counts, h-index, or citing works are available for these brand-new preprints. This axis would measure scholarly influence through citations and downstream adoption, and cannot be estimated without aggregated citation coverage.
Focus0/100
No score yet: the four papers share one theme, but field labels are missing, so topical concentration cannot be computed. This axis would measure how narrowly or broadly the work spans fields, venues, and keywords.
Collaboration44/100
Score 44/100, from raw data showing 8 distinct frequent coauthors: Mingxuan Wang, Fei Luo, Bo Wang, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, and Jungong Han. All four catalogued 2026 papers place Yao in a middle author position with no corresponding-author record. Together this points to a compact, recurring co-author pool and a contributing-team role rather than an independent or cross-institution collaboration pattern.
Activity—
No score yet: all four entries carry the same 2026 year, so no multi-year cadence or recent-momentum trend can be computed. This axis would measure publication rhythm — outputs per year and gaps — once earlier or later records are linked.
Reproducibility46/100
No score yet: none of the four papers carries a reproReadiness assessment, code, or artifact record. This axis would measure whether releases include code, data, and documentation sufficient for independent replication.
Research summary
Guorun Yao is a researcher working on memory compression and control for long-horizon LLM agents. The profile is name-resolved from public bibliographic data: no institution, country, or ORCID is on file, so identity should be read with care. The evidence base is a tight cluster of four 2026 arXiv preprints in the Evolve catalog that approach one problem from complementary angles — making long-horizon agents efficient by deciding what to keep in memory. StateComp learns when to compress agent history; DRSR learns set-level deletion risk; 'Memory Control Signals Emerge Before Action in Long Horizon Agents' reports that memory-control signals precede action; and 'Stable Geometry with Divergent Task Evidence for Efficient Long-Horizon Agent Compression' targets stable representations when task evidence diverges. Across all four, Yao sits in a middle author position with no corresponding-author record. The collaboration dimension scores 44/100, grounded in 8 distinct frequent coauthors (Mingxuan Wang, Fei Luo, Bo Wang, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, Jungong Han), which fits a steady contributor inside a compact team. Because the preprints are new, citations, multi-year cadence, and reproducibility signals are not yet measurable; those axes are reported as missing rather than estimated. This is an early-stage profile concentrated on one active research front.
Collaboration pattern
All four catalogued 2026 papers list Yao in a middle author position with no corresponding-author role. The raw record shows 8 distinct frequent coauthors — Mingxuan Wang, Fei Luo, Bo Wang, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, and Jungong Han — and the collaboration dimension scores 44/100. This reads as a compact, recurring co-author pool: Yao appears to work within a stable team on one shared agenda rather than rotating across many groups.
Research keywords
Highlights
StateComp: learning when to compress agent history
The 2026 preprint 'StateComp: Learning When to Compress History in Long Horizon Agents' addresses history compression as a learned decision — when to compress, not just what — within long-horizon agents. Yao is a middle author.
DRSR: set-level deletion risk for leaner agents
'DRSR: Learning Set-Level Deletion Risk for Efficient Long-Horizon Agents' (2026) learns deletion risk at the level of content sets rather than single items, targeting efficiency in long-horizon agents. Yao is a middle author.
Memory-control signals emerge before action
'Memory Control Signals Emerge Before Action in Long Horizon Agents' (2026) reports, in its title-level finding, that memory-control signals arise ahead of agent actions — an empirical cue relevant to compression design. Yao is a middle author.
Stable geometry under divergent task evidence
'Stable Geometry with Divergent Task Evidence for Efficient Long-Horizon Agent Compression' (2026) pursues stable representations when task evidence diverges, so compression remains reliable across long horizons. Yao is a middle author.
Timeline
- 2026'StateComp: Learning When to Compress History in Long Horizon Agents' appears in the Evolve catalog as an arXiv preprint; Yao is a middle author.
- 2026'DRSR: Learning Set-Level Deletion Risk for Efficient Long-Horizon Agents' appears on arXiv, learning deletion risk at the set level.
- 2026'Memory Control Signals Emerge Before Action in Long Horizon Agents' appears on arXiv as the empirical companion in the cluster.
- 2026'Stable Geometry with Divergent Task Evidence for Efficient Long-Horizon Agent Compression' appears on arXiv as a further catalog entry.
- 2026Across the year, all four preprints share one agenda — efficient long-horizon agents via history and memory compression — with 8 distinct coauthors and middle authorship throughout.
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.
- Identity is name-resolved from public bibliographic data; no institution, country, or ORCID is on file, and 'Guorun Yao' may be ambiguous.
- All bibliometric signals come from four 2026 arXiv preprints in the Evolve catalog; no peer-reviewed venue, citation counts, or h-index are available.
- Only the collaboration dimension has sufficient data for a score (44/100, from 8 distinct coauthors); output, impact, focus, activity, and reproducibility are unscored.
- No Evolve reproducibility verdicts (reproReadiness) or field labels exist for any paper, and paper slugs are not yet assigned.