Long-horizon agent efficiency: middle-position co-author on four 2026 arXiv preprints on agent memory compression
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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—
No output data is available yet: this axis would measure publication volume over time, venue types, and authorship roles across the full record. The four arXiv preprints in the Evolve catalog are the only output signal currently visible.
Impact—
Impact cannot be scored yet: this axis would measure citation counts, h-index, and uptake of the four 2026 preprints. The current snapshot contains no citation or venue data, so no estimate is offered.
Focus0/100
Focus is not yet scored: this axis would measure topical concentration across the full publication record. The four catalog papers all concern long-horizon agent compression, but the sample is too small and too recent to quantify.
Collaboration53/100
The collaboration dimension scores 44 out of 100, computed from 8 distinct coauthors in the underlying bibliometric data. The recurring partner set includes Mingxuan Wang, Fei Luo, Guorun Yao, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, and Jungong Han. On all four catalog papers Bo Wang sits in a middle author position, neither first nor corresponding, indicating contribution within a team rather than independent leadership. The moderate score reflects a small but stable collaborator pool.
Activity—
Activity data is insufficient: this axis would measure publication cadence, recency, and year-over-year rhythm. Only four preprints, all dated 2026, are visible, so no sustained activity pattern can be established.
Reproducibility46/100
Reproducibility is not yet assessed: this axis would measure code and artifact release, benchmark availability, and Evolve repro-readiness checks. None of the four catalog papers carries repro-readiness data in this snapshot.
Research summary
Bo Wang is an early-stage profile in the Evolve catalog, name-resolved from public bibliographic data and currently documented through four 2026 arXiv preprints rather than an institutional record. All four papers address a single technical theme: making long-horizon LLM agents efficient by deciding what to keep, compress, or delete from their interaction history. StateComp learns when to compress history; DRSR learns set-level deletion risk; the stable-geometry paper examines geometric stability under divergent task evidence during compression; and the memory-control-signals paper analyzes when memory-related control signals emerge relative to action in agent trajectories. Bo Wang appears as a middle-position author on all four, neither first nor corresponding author, consistent with a contributing-researcher role within a team. The collaboration dimension scores 44 out of 100 on a base of 8 distinct coauthors, including recurring collaborators such as Mingxuan Wang, Fei Luo, and Jungong Han. Output, impact, focus, activity, and reproducibility dimensions currently lack sufficient underlying data and are reported as missing rather than estimated, so the profile should be read as an early-stage, preprint-era snapshot.
Collaboration pattern
Bo Wang's collaboration footprint in this snapshot covers 8 distinct coauthors, with Mingxuan Wang, Fei Luo, Guorun Yao, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, and Jungong Han as recurring partners. The collaboration dimension scores 44 out of 100. On all four catalog papers Bo Wang is a middle-position author and not a corresponding author, consistent with contributing to a shared agenda on long-horizon agent compression rather than leading it.
Research keywords
Highlights
StateComp: learning when to compress agent history
A 2026 arXiv preprint proposing a learned method for deciding when to compress interaction history in long-horizon agents, targeting efficiency over extended task trajectories.
Set-level deletion risk for efficient long-horizon agents (DRSR)
Frames history pruning as set-level deletion-risk learning, aiming to keep long-horizon agents efficient while limiting the information lost by removing past states.
Stable compression geometry under divergent task evidence
A 2026 preprint studying geometric stability conditions for compressing long-horizon agent state when task evidence diverges, tying stable representations to efficient agent execution.
Memory control signals emerge before action
An analysis of long-horizon agents showing that control signals related to memory arise before action, informing when and how history compression should be triggered.
Timeline
- 2026StateComp, a learned history-compression method for long-horizon agents, appears on arXiv.
- 2026DRSR introduces set-level deletion-risk learning for efficient long-horizon agents.
- 2026A preprint on stable compression geometry under divergent task evidence is released.
- 2026An analysis showing memory control signals emerge before action in long-horizon agents is released.
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.
- Name-resolved from public bibliographic data only; 'Bo Wang' is a common name and no ORCID or institutional affiliation is on record, so identity disambiguation carries uncertainty.
- Bibliometric scores come from a single snapshot; no citation counts, h-index, venue, or peer-review data are available.
- All four catalog papers are 2026 arXiv preprints with no field labels, evolveVerdict, or repro-readiness data attached, and no paper slugs exist to deep-link highlights.
- Five of six radar dimensions (output, impact, focus, activity, reproducibility) lack sufficient underlying data and are reported as missing, not scored.