Alibaba Group ML researcher: agentic RL scheduling, LoRA fine-tuning of LLM agents, and multimodal model merging
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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—
The output score is not computed yet: the OpenAlex record is too thin and possibly name-ambiguous to score reliably. This axis would measure publication volume and consistency over time; the only visible signals are 10 recent-3-year works and sparse yearly counts since 2023.
Impact—
The impact score is not computed yet: citation signals are essentially absent (h-index 1, i10-index 0), which is too thin to score. This axis would measure citation uptake and scholarly influence of the work across fields and years.
Focus0/100
Focus scores 0/100, reflecting a highly dispersed record. The current field distribution splits evenly — one third each for reinforcement learning for LLM agents, parameter-efficient fine-tuning of LLM agents, and model merging for multimodal LLMs — while OpenAlex topics still span robotics RL (6), domain adaptation and few-shot learning (4), path planning (3), and image enhancement/processing (3 each).
Collaboration55/100
Collaboration scores 55/100, based on 15 distinct coauthors in the last three years. On all three cataloged 2026 papers Xue sits in a middle author position without corresponding role, indicating contribution within a team rather than leading it. Recurring names include Yinggui Wang and Jinkui Ren, alongside Xiaobing Tu, Xiantao Zhang, and Debin Zhao.
Activity70/100
Activity scores 70/100, supported by 10 works in the last three years and a rising yearly trajectory: 1 work in 2023, 3 in 2024, 3 in 2025, and 4 in 2026. Three arXiv preprints already appear in 2026, all reviewed by Evolve, indicating a currently intense and accelerating output cadence.
Reproducibility32/100
Reproducibility scores 32/100, the mean readiness across the 3 papers Evolve reviewed. HarnessBandit is rated medium, while TAC-Merge and BQ-LoRA are rated low — both dinged for no code release. The pattern suggests solid method write-ups but limited artifact sharing, so independent replication currently requires substantial reimplementation.
Research summary
Rongkun Xue is a machine learning researcher at Alibaba Group (China) whose Evolve profile is name-resolved from OpenAlex bibliographic data (no ORCID linked). The current record is early-stage: h-index 1, i10-index 0, and 10 works in the last three years, with yearly counts rising from 1 in 2023 to 3 in 2024, 3 in 2025, and 4 in 2026. The recent focus has shifted decisively toward large-language-model methods: all three papers Evolve has reviewed are 2026 arXiv preprints covering reinforcement learning for LLM agents (HarnessBandit), parameter-efficient fine-tuning of LLM agents (BQ-LoRA), and model merging for multimodal LLMs (TAC-Merge). This is a sharp pivot from the historical OpenAlex topic mix, which was weighted toward reinforcement learning in robotics (6 topic-assigned works), domain adaptation and few-shot learning (4), robotic path planning (3), and image enhancement and processing (3 each). Field distribution is evenly split across the three current ML subfields (one third each), yielding a focus score of 0. On the cataloged work, Xue appears as a middle author without corresponding role, collaborating within a group of 15 distinct recent coauthors. Reproducibility readiness across the three reviewed papers averages 32/100, held back by missing code releases.
Collaboration pattern
Xue's recent work is team-based: 15 distinct coauthors in the last three years, with Yinggui Wang and Jinkui Ren recurring most often in the frequent-collaborator list, alongside Xiaobing Tu, Zhengxi Liu, Xiantao Zhang, Haochen Li, Miancan Liu, Ziyuan Chen, Hongliang Wei, and Debin Zhao. On all three cataloged 2026 papers Xue is a middle author and not the corresponding author, consistent with a contributing researcher role within an Alibaba Group ML team.
Research keywords
Highlights
HarnessBandit: bandit-scheduled harness selection for agentic RL
TAC-Merge traces cross-layer influence for multimodal model merging
BQ-LoRA compresses agent trajectory updates under fixed LoRA rank
Timeline
- 2026BQ-LoRA preprint appears on arXiv, introducing quotient-geometry LoRA adaptation for LLM agents; Evolve reviews it as solid math with modest gains and no code.
- 2026TAC-Merge preprint appears on arXiv, tracing cross-layer multimodal influence with curvature graphs for model merging; Evolve notes solid gains but no code release.
- 2026HarnessBandit preprint appears on arXiv, scheduling training harnesses per GRPO step for multi-harness agentic RL; Evolve rates its reproducibility 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.
- Identity is resolved from OpenAlex only (no ORCID linked); 'Rongkun Xue' is not a rare name, and the record may mix namesakes, which could also explain the sparse pre-2023 yearly counts.
- All bibliometrics come from the provided OpenAlex snapshot; no output or impact score could be computed from the thin citation record (h-index 1, i10-index 0).
- The reproducibility rating covers only the 3 Evolve-reviewed arXiv preprints; none is a confirmed peer-reviewed publication.
- Affiliation and country are taken as given in the source record and have not been independently verified.