Tencent ML researcher in reinforcement learning, systematic generalization, and role-playing agent training.
3
papers here
3
reviewed by Evolve
0
first-author
1
corresponding
—
works · OpenAlex
17
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—
Not yet scored: publication-volume data is incomplete in the current snapshot. This axis would measure cumulative works and yearly productivity; the 57 works logged in the last three years hint at momentum but do not yet support a score.
Impact—
Not yet scored: citation coverage is incomplete for this record. This axis would capture aggregate citations, h-index reach, and uptake in follow-up work; although an h-index of 17 is listed, the platform cannot yet compute a reliable impact score.
Focus0/100
Scores 100 on focus. OpenAlex topic counts concentrate on reinforcement learning in robotics (35 works), with supporting clusters in topic modeling (15), multimodal machine learning applications (15), and AI in games (14); recent cataloged papers extend this into role-playing agents and systematic-generalization benchmarks. The record's field distribution is dominated by machine learning (neural network generalization), indicating a coherent research identity across applied RL and generalization analysis.
Collaboration63/100
Collaboration scores 44, the lowest of the computed axes. The record lists 8 distinct coauthors, with repeated collaborators including Zheng Zhang, Liu Liu, Qi Chai, Peilin Zhao, and Yatao Bian. Both cataloged 2026 papers place Deheng Ye in a middle author position, consistent with a contributor role inside small-to-medium teams rather than a large-network hub or habitual corresponding author.
Activity100/100
Activity scores 100. The OpenAlex record logs 57 works in the last three years, and per-year counts climb from 10 (2021) and 9 (2022) to 12 (2023), 12 (2024), and a peak of 28 (2025), with 17 already in 2026. This steep recent ramp marks the current period as the most productive stretch in a record reaching back to 2014.
Reproducibility58/100
Reproducibility scores 100, computed from 1 assessed paper with a mean readiness of 100. Evolve rates the TranSGrid systematic-generalization study as high reproducibility — controlled and convincing — though limited to small Transformers without error bars. The AdvRole curriculum paper is rated medium, with modest gains and no significance testing. With only one assessed paper, this perfect score should be read as a narrow but positive signal.
Research summary
Deheng Ye is a machine learning researcher at Tencent (China) whose OpenAlex-resolved record spans 2014–2026, with an h-index of 17 and an i10-index of 27. The profile is strongest on applied and analytical reinforcement learning: OpenAlex topic counts lead with reinforcement learning in robotics (35 works), followed by topic modeling (15), multimodal machine learning applications (15), and artificial intelligence in games (14), with software engineering research (11) also represented. Activity is the standout dimension: 57 works in the last three years and a yearly ramp from 10 works in 2021 to 28 in 2025, with 17 logged in 2026 — the most productive stretch of the record. Two 2026 arXiv papers sit in the Evolve catalog, both in middle author positions. TranSGrid interrogates whether current systematic-generalization benchmarks actually test deductive, inductive, and abductive reasoning, earning a high reproducibility rating; AdvRole builds an adversarial closed-loop curriculum for role-playing agents with GRPO-based self-play, rated medium, with modest gains and no significance testing. Collaboration is the weakest computed axis (44), built on 8 distinct coauthors including Zheng Zhang, Peilin Zhao, and Yatao Bian. Output and impact dimensions are not yet scored because publication-volume and citation coverage are incomplete; the numbers above come solely from the provided snapshot.
Collaboration pattern
Collaboration is the lowest computed axis (44). The record lists 8 distinct coauthors, and repeated partners include Zheng Zhang, Liu Liu, Qi Chai, Peilin Zhao, Mao Zheng, Hao Wang, Chengwen Qi, and Yatao Bian. Both cataloged 2026 papers place Deheng Ye in a middle author position without corresponding status, consistent with contributing specialist roles within Tencent research teams rather than solo or corresponding-author leadership.
Research keywords
Highlights
TranSGrid: a reasoning-centered audit of systematic generalization benchmarks
AdvRole: adversarial closed-loop curricula for role-playing agents
2025 is the record's most productive year
Per-year output climbs from 10 works in 2021 to 28 in 2025, with 17 already logged in 2026 and 57 works across the last three years — a steep recent ramp on the OpenAlex record.
Timeline
- 2014First year in the OpenAlex record, with 1 work — the start of a publication trail running through 2026.
- 2016An early peak of 7 works, the highest single-year count of the 2014–2019 stretch.
- 2021Output climbs to 10 works, launching a sustained multi-year run through the mid-2020s.
- 2025Most productive year on record with 28 works, part of 57 works across the last three years.
- 2026Two papers enter the Evolve catalog — AdvRole on role-playing agent curricula and TranSGrid on systematic generalization — with 17 works already logged this year.
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 a single OpenAlex identity record and may reflect incomplete coverage or name-variant merging.
- Output and impact dimensions are unscored in this snapshot because publication-volume and citation coverage are insufficient; no estimates are given.
- The reproducibility score (100) is based on only 1 assessed paper and should be read as a narrow signal.
- 'Deheng Ye' is name-resolved from public bibliographic data and the listed ORCID (0000-0002-1754-1837); the profile is limited to the provided snapshot.