Shanghai Academy of Social Sciences LLM-agent researcher building agent environments and self-improving scientific AI
5
papers here
5
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
7
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
This axis would measure total publication volume and productivity against field norms, but the scoring inputs are not available in this snapshot; the year-by-year record is sparse with multi-year gaps, so no output score is reported.
Impact—
Impact scoring needs citation counts and field-normalized citation percentiles, which are not provided here; the only citation-side figures present are h-index 7 and i10-index 6, too limited to score influence.
Focus0/100
The focus score is 0, the maximal-dispersion signal, because this identity's OpenAlex topic profile is led by biomedical themes — Immune cells in cancer (17 works), Cancer Immunotherapy and Biomarkers (15), Ferroptosis and cancer prognosis (9) and single-cell/spatial transcriptomics (8) — while the field distribution and all three reviewed papers sit entirely in AI. This split suggests name conflation or a sharp domain shift and warrants caution.
Collaboration88/100
Collaboration scores 82, backed by 59 distinct coauthors in the recent three-year window across 30 works — a wide, team-oriented network. Twelve recurring collaborators are recorded, including Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Koutian Wu and Zihan Zhou, concentrated around the 2026 arXiv agent papers. On the three catalogued papers Wu holds middle or last authorship, with no corresponding-author role flagged.
Activity95/100
Activity is scored at the maximum of 100: 30 works in the last three years, with the indexed year counts accelerating from 14 in 2023 to 8 in 2025 and 19 already in 2026. All three Evolve-reviewed papers appeared in 2026, confirming that current output is concentrated in arXiv agent research. This is a very high tempo for an early-stage profile.
Reproducibility41/100
Reproducibility averages 43 across three assessed papers. Evolve rates ScienceIDE and ScienceBuddy at medium readiness — both report real gains (eval/SFT/RL improvements; 42.2% to 73.3% over three cycles) but with narrow coverage and single-run, simulated-user evaluation — while the Discovery Foundation Models paper is rated low since nothing is implemented or measured. Systems work looks reproducible; the conceptual paper is unverified.
Research summary
Yingcheng Wu, affiliated with the Shanghai Academy of Social Sciences (China), works on large language model agents for scientific research. This profile is name-resolved from OpenAlex bibliographic data, as no ORCID is on record. The Evolve catalog reviews three 2026 arXiv papers. ScienceIDE converts scientific codebases into calibrated, agent-learnable environments and reports real evaluation, supervised fine-tuning and reinforcement-learning gains, though Evolve notes narrow coverage and selectively reported benchmark results. ScienceBuddy introduces recursive-in-recursive self-improvement, combining harness evolution with rubric-reward GRPO training to lift a scientific agent from 42.2% to 73.3% across three cycles, with small, single-run, simulated-user evaluation. A third, last-author position paper defines Discovery Foundation Models, an unimplemented conceptual framework. The bibliometric radar scores activity at the maximum 100 (30 works in the last three years; 19 works indexed in 2026) and collaboration at 82 (59 distinct coauthors); reproducibility averages 43 across the three assessed papers. Output and impact cannot be scored: the only citation-side figures are h-index 7 and i10-index 6. Notably, this identity's OpenAlex topics are biomedical-led — immune cells in cancer (17), cancer immunotherapy and biomarkers (15), ferroptosis (9) — while its field distribution is entirely AI, a split that lowers the focus score to 0 and suggests identity conflation or a rapid domain transition.
Collaboration pattern
Collaboration is broad and team-based: 59 distinct coauthors over the last three years across 30 works, with a stable core of 12 recurring collaborators including Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Koutian Wu, Zihan Zhou and Yuanbo Pang. On the three catalogued 2026 arXiv papers Wu appears in middle or last authorship, none as corresponding author, consistent with contributing inside a larger LLM-agents group rather than leading an independent lab.
Research keywords
Highlights
ScienceIDE turns scientific codebases into agent-learnable environments
ScienceBuddy lifts a scientific agent from 42.2% to 73.3% via recursive self-improvement
Discovery Foundation Models: an ambitious framework, not yet evidence
Timeline
- 2026Published ScienceIDE on arXiv, converting the world's scientific codebase into agent-learnable environments with evaluation, SFT and RL gains.
- 2026Released ScienceBuddy, a recursive-in-recursive self-improvement method raising a scientific agent from 42.2% to 73.3% over three cycles.
- 2026Co-authored the Discovery Foundation Models position paper, proposing a taxonomy and process protocol for open-ended discovery intelligence.
Papers in the Evolve catalog
JEPA-Anything: Learning Predictive Models across Different Worlds
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents
Discovery Foundation Models: Toward Open-Ended Discovery Intelligence
An immune world model for multiscale forecasting and therapeutic hypothesis generation
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 are derived from OpenAlex only; no ORCID is on record, so this profile is name-resolved from public bibliographic data.
- OpenAlex topics for this identity are biomedical-dominated while the reviewed papers and field distribution are AI-only — likely identity conflation or mixed records; cross-topic figures should be read with caution.
- The year-by-year publication counts contain long gaps (2010-2019 and 2021-2022 are absent), so long-run productivity trends cannot be established.
- All three reviewed papers are 2026 arXiv preprints; no peer-reviewed venue information is within the current data scope.