Early-stage Zhejiang University researcher in data-efficient language modeling and autonomous AI agent evaluation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 score yet: OpenAlex productivity aggregates for this identity are too thin to score. This axis would measure publication volume and rate — total works and per-year output — beyond the 4 recent-window works and h-index 0 currently recorded.
Impact—
Impact is unscored: citation-based influence signals have not accumulated in this snapshot (h-index 0, i10-index 0). This axis would measure citation counts, citing works, and field-weighted influence once the 2026-only record gathers uptake.
Focus0/100
Focus scores 0, reflecting an evenly split rather than concentrated record: Evolve's field distribution assigns 0.5 to data-efficient language model training and 0.5 to evaluation of autonomous scientific research systems, while OpenAlex topics span five areas (advanced combinatorial mathematics x2, topic modeling, NLP techniques, complexity and algorithms in graphs, AI-based problem solving and planning). Work so far covers two AI tracks plus mathematics topics rather than one core line.
Collaboration63/100
Collaboration scores 63, grounded in 23 distinct coauthors in the OpenAlex snapshot and 12 recurring collaborators, including Kaihao Zhu, Shenzhan Hong, Zichen Li, and Hongsheng Chen. Both Evolve-reviewed 2026 papers list this researcher in a middle author position without corresponding authorship, consistent with large-team benchmark and engineering projects rather than solo-led work.
Activity47/100
Activity scores 47. The OpenAlex snapshot records 4 works inside the recent three-year window, all attributed to 2026 in countsByYear, with no earlier-year works shown. This thin per-year footprint keeps the score moderate despite clearly active 2026 output on arXiv.
Reproducibility78/100
Reproducibility scores 78, the mean readiness Evolve assigned across 2 reviewed papers. The BabyLM paper rates high, with careful controls but gains hinging on a 203-item near-chance metric and 2 seeds; the Qiushi Engine paper rates medium, with disciplined self-evaluation while attribution and judge validity rest on self-reported evidence.
Research summary
Taowen Deng is a researcher at Zhejiang University, China; this profile is name-resolved from public OpenAlex bibliographic data. The snapshot shows a very early citation record — h-index 0 and i10-index 0 — but clearly active 2026 output: 4 works fall in the recent three-year window, produced with 23 distinct coauthors. OpenAlex topics on the record span advanced combinatorial mathematics (2), topic modeling, natural language processing techniques, complexity and algorithms in graphs, and AI-based problem solving and planning, while Evolve's field distribution splits evenly at 0.5 for data-efficient language model training and 0.5 for evaluation of autonomous scientific research systems — a two-track profile. Two Evolve-reviewed arXiv papers anchor it. A middle-position contribution to 'Data-Efficient Language Modeling: From Frontier Advancement to Principle-Guided Model Improvement' reports an autonomous BabyLM program topping the 8 Sep 2026 Strict-Small snapshot at 42.25, with careful controls but gains hinging on a 203-item near-chance metric and 2 seeds. A second paper reports 0.816 on AstaBench E2E-Bench-Hard for the Qiushi Engine, a 3.3x best perfect-task rate, though attribution and judge validity rest on self-reported evidence. Evolve's reproducibility review covers both papers at a mean readiness of 78; radar scoring for output and impact remains pending for insufficient data.
Collaboration pattern
Taowen Deng works in sizeable teams: the OpenAlex snapshot lists 23 distinct coauthors, with 12 recurring collaborators such as Kaihao Zhu, Shenzhan Hong, Zichen Li, Jincheng Mi, and Hongsheng Chen. On both Evolve-reviewed 2026 papers this researcher holds a middle author position and is not the corresponding author, suggesting contribution within large, engineering-oriented groups working on data-efficient language modeling and autonomous-agent evaluation at Zhejiang University.
Research keywords
Highlights
Autonomous BabyLM program tops Strict-Small snapshot at 42.25
Qiushi Engine scores 0.816 on AstaBench E2E-Bench-Hard
Timeline
- 2026Middle-position author on 'Data-Efficient Language Modeling: From Frontier Advancement to Principle-Guided Model Improvement' (arXiv), whose autonomous BabyLM program topped the 8 Sep 2026 Strict-Small snapshot at 42.25.
- 2026Co-authored 'Qiushi Engine on AstaBench E2E-Bench-Hard' (arXiv), reporting 0.816 on E2E-Bench-Hard and a 3.3x best perfect-task rate.
- 2026OpenAlex snapshot attributes all 4 works in the recent three-year window to 2026, co-written with 23 distinct coauthors.
- 2026Evolve completes reproducibility review of the 2 cataloged papers, with a mean readiness score of 78 (high and medium).
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.
- Bibliometrics are name-resolved from OpenAlex only; no ORCID is linked, so the record may mix homonyms — the combinatorial-mathematics topics may not belong to the same person as the NLP and agent work.
- Citation aggregates are effectively absent: h-index and i10-index are 0, and countsByYear covers only 2026 (4 works); radar scores for output and impact are missing due to insufficient data.
- Both cataloged papers are 2026 arXiv preprints reviewed by Evolve; headline numbers (42.25 on BabyLM Strict-Small; 0.816 on AstaBench E2E-Bench-Hard) are benchmark reports with noted reliance on a 203-item near-chance metric, 2 seeds, and self-reported evidence.
- Career facts such as position title, degrees, and awards are outside the input data and are therefore not described.