Machine learning researcher at Nanjing University spanning RL for LLM reasoning and offline data-driven optimization.
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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 metrics are not yet available: the underlying record lacks reliable totals of publications and productivity aggregates, so no score is given. This axis would measure cumulative works, yearly output trends, and authorship roles.
Impact—
Impact data is insufficient for scoring: citation counts and influence signals are not yet available in the underlying record, so this axis is left unscored. It would measure citations, yearly citation trajectory, and uptake of the work by other researchers.
Focus100/100
Focus scores a perfect 100 because the record is tightly concentrated: OpenAlex assigns the reviewed activity to a single field, machine learning for large language models, while topic mass clusters on optimization — multi-objective (5 works), metaheuristics (4), evolutionary algorithms (3), and bandit algorithms (3) — a coherent optimization-plus-RL line rather than scattered interests.
Collaboration54/100
Collaboration scores 54, based on 14 distinct coauthors across 7 works in the last three years — a moderate, team-embedded pattern. The recurring partner list includes Chao Qian, Ke Xue, Chen-Guang Wang, Yizhuo Li, Ganqu Cui, and Ning Ding, suggesting work inside organized research groups rather than solo authorship.
Activity61/100
Activity scores 61 on the strength of 7 works over the last three years, with the yearly distribution heavily weighted to 2026 (6 works) against 1 in 2024 — an accelerating cadence consistent with an early-career researcher whose arXiv preprints are appearing now. Sustained multi-year output would lift this score further.
Reproducibility55/100
Reproducibility scores 55, computed from 1 reproducibility-assessed paper with a mean readiness of 55. Both reviewed preprints are rated medium: the PPO critic-learning study brings strong diagnostics and ablations, but its evidence rests on single-seed runs and terminal rewards only, while the offline-optimization work flags its theory-to-practice link as heuristic.
Research summary
Rong-Xi Tan is an early-career researcher at Nanjing University whose profile, name-resolved from public OpenAlex bibliographic data, centers on optimization and reinforcement learning. The OpenAlex snapshot records 7 works in the last three years, an h-index of 1, and 14 distinct coauthors, with topic mass concentrated in multi-objective optimization (5), metaheuristic optimization (4), evolutionary algorithms (3), and bandit algorithms (3). Two 2026 arXiv preprints reviewed by Evolve define the current research front. The first, Rethinking Critic Learning in PPO, identifies and mitigates value flattening in PPO critics, showing that SP3O's 3-anchor sparse supervision helps, with strong diagnostics and ablations though the evidence is single-seed and terminal-reward-only. The second, Rethinking Learnability in Offline Data-driven Optimization, shows trajectory-limited learnability suffices for offline optimization in three settings and introduces UGTL, ranked 3.1/25 on Design-Bench, while acknowledging the theory-to-practice link remains heuristic; both carry medium reproducibility readiness. Radar scores show perfect focus (100) with moderate activity (61), collaboration (54), and reproducibility (55), while output and impact lack sufficient data. Positioned as a middle author and not corresponding on either reviewed paper, Tan works within a network including Chao Qian, Ke Xue, Chen-Guang Wang, Ganqu Cui, and Ning Ding.
Collaboration pattern
Tan publishes as a middle author within a dense co-author network: 14 distinct collaborators over the last three years, with a recurring core spanning Chao Qian, Ke Xue, Chen-Guang Wang, Yizhuo Li, Jianhao Yan, Ganqu Cui, and Ning Ding. Neither reviewed paper lists Tan as corresponding author, consistent with a contributor role inside organized optimization and LLM research groups.
Research keywords
Highlights
Identifying and mitigating value flattening in PPO critics
Trajectory-limited learnability for offline data-driven optimization
Timeline
- 2024First visible activity in the OpenAlex record: 1 work counted in this year, marking the start of the indexed publication trail.
- 2026Rethinking Critic Learning in PPO posted to arXiv, introducing the value-flattening diagnosis and the SP3O 3-anchor sparse supervision analysis.
- 2026Rethinking Learnability in Offline Data-driven Optimization posted to arXiv, with UGTL ranked 3.1/25 on Design-Bench.
- 2026Yearly output peaks at 6 works in the OpenAlex counts, with the co-author network reaching 14 distinct collaborators over three years.
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 an OpenAlex snapshot (identity: openalex); the output and impact axes lack sufficient underlying data and are unscored.
- Only 2 papers have been reviewed by Evolve; the profile reflects that reviewed subset plus coarse OpenAlex counts, not a full publication audit.
- The h-index (1) and i10-index (0) reflect a very early-career record and should be re-read as the record matures.
- Identity is name-resolved from public bibliographic data linked to ORCID 0009-0005-1547-051X; affiliation and disambiguation follow the index.