Fudan University researcher working on NLP and RL for LLM agents: context compression, multi-turn policy optimization.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Output data is currently insufficient — the OpenAlex record shows h-index 0 and i10-index 0, suggesting incomplete publication linkage. This axis would measure total works, venue spread, and yearly productivity once the record is fully linked.
Impact—
Impact data is insufficient: h-index and i10-index are both 0 in the OpenAlex record and no citation counts are provided, so influence cannot be scored. This axis would measure citations, citing works, and field-weighted influence as citation data accumulates.
Focus0/100
The focus score is 0, consistent with a deliberately broad portfolio. The field distribution splits evenly — 0.5 natural language processing versus 0.5 reinforcement learning for LLM agents — and topic clusters reach across five areas: multimodal ML applications (5), topic modeling (4), RL in robotics (3), neural network applications (2), and NLP techniques (2). The two reviewed papers sit on opposite sides of this 50/50 split.
Collaboration41/100
Collaboration scores 41 on the 0–100 scale. The recent three-year window contains 8 works with 7 distinct coauthors — a compact circle rather than a large network. Evolve lists seven recurring collaborators: Yuyao Zheng, Haipeng Sun, Junwei Bao, Lemao Liu, Hongfei Jiang, Yang Song, and Yifan Wang. On both reviewed 2026 papers Dou appears as last author without corresponding-author flags, consistent with senior supervision of small teams.
Activity64/100
Activity scores 64. The OpenAlex counts-by-year snapshot attributes 8 works to 2026, and the recent three-year window likewise reports 8 works — essentially all recent output concentrated in the current year. Both cataloged papers are 2026 arXiv releases, matching this burst pattern. Earlier-year counts are absent from the snapshot, so the pre-2026 cadence cannot be characterized from this data.
Reproducibility55/100
Reproducibility scores 55, computed from 2 assessed papers with a reproducibility mean of 55. Both reviewed works carry a medium readiness rating: PGPO's gains are three-seeded but small-margin on only two benchmarks, while the machine-interpretable information paper claims roughly 7% FLOPs against full context yet with lagging F1 and thin evidence. Reproducibility is plausible but would require careful re-verification.
Research summary
Dejing Dou is a researcher at Fudan University whose OpenAlex record splits evenly between two fields: natural language processing and machine learning — reinforcement learning for LLM agents (0.5 versus 0.5 in the field distribution). Evolve's reviewed catalog contains two 2026 arXiv papers, both with Dou as last author. PGPO introduces potential-guided, step-level credit assignment into group policy optimization (GRPO-style) for multi-turn agentic tasks; Evolve rates the gains largely credible but small-margin, three-seeded, and demonstrated on only two benchmarks. The second line, machine-interpretable information, compiles documents into compact, searchable, readable protocol states — 56-token states consumable by frozen LLMs — with Residual-MII beating full-context exact match on HotpotQA at roughly 7% of the FLOPs, though F1 lags and evidence remains thin. The record's topic clusters reach across five areas — multimodal machine learning applications (5), topic modeling (4), reinforcement learning in robotics (3), advanced neural network applications (2), and NLP techniques (2) — signaling breadth beyond the cataloged fields. Recent activity shows 8 works in the three-year window, all attributed to 2026, with 7 distinct coauthors. Reproducibility averages 55 across 2 assessed papers, both rated medium readiness. Citation-based output and impact scores are unavailable, so this profile remains early-stage.
Collaboration pattern
The collaboration footprint is compact: 8 works in the recent three-year window with 7 distinct coauthors, scoring 41 on the 0–100 radar. Seven recurring collaborators are listed — Yuyao Zheng, Haipeng Sun, Junwei Bao, Lemao Liu, Hongfei Jiang, Yang Song, and Yifan Wang. On both reviewed 2026 arXiv papers Dou appears as last author without a corresponding-author flag, a position consistent with senior supervision of small, stable teams rather than large distributed consortia.
Research keywords
Highlights
PGPO: potential-guided step credit for multi-turn LLM agents
Compiling documents into machine-interpretable protocol states
Timeline
- 2026PGPO: Potential-Guided Policy Optimization for Multi-Turn Agentic Tasks appears on arXiv; Dou is last author.
- 2026Machine-Interpretable Information paper (compiling documents into searchable, readable protocol states) is released on arXiv; Dou is last author.
- 2026The OpenAlex counts-by-year snapshot attributes 8 works to 2026, matching the 8 works reported in the recent three-year window.
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 figures come from the OpenAlex record only; no ORCID is linked, and h-index/i10-index show 0, likely reflecting incomplete publication linkage.
- Output and impact dimensions are unscored because citation-based data is insufficient in the snapshot.
- Qualitative assessments draw on only two Evolve-reviewed papers, both 2026 arXiv preprints; no peer-reviewed venue records appear in the catalog.
- This profile is name-resolved from public bibliographic data; the name may correspond to multiple individuals, and the Fudan University affiliation follows the OpenAlex record.