Tencent researcher on efficient LLM reasoning: latent steering, rotary position embeddings, reinforcement finetuning
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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—
Not yet scored: this axis would measure publication volume and consistency over time, such as yearly paper counts and venue spread. The snapshot provides no such counts, so no estimate is made.
Impact—
Not yet scored: this axis would capture citation-based influence, including total citations and h-index-style aggregates. No citation data exists in the provided snapshot, so the axis stays unscored rather than guessed.
Focus100/100
Scored 100/100: the snapshot's field distribution lists exactly one field — natural language processing (LLM reasoning) — so all field-classified catalog work falls in a single area. The three 2026 papers (latent steering for reasoning efficiency, higher-dimensional rotary position embeddings, and dynamic example mining for reinforcement finetuning) all sit inside the large-language-model space, matching a narrow and coherent specialization rather than a spread across disciplines.
Collaboration66/100
Scored 66/100 from 27 distinct coauthors in the snapshot, 12 of whom recur across papers — including Samm Sun, Haoru Tan, Sitong Wu, and Yanfeng Chen — indicating a stable core team within a moderately wide network. On all three cataloged papers Xie appears in a middle author position and is not listed as corresponding author, suggesting a contributor role within group-led projects.
Activity—
Not yet scored: this axis would track publishing cadence across years, including gaps and recent output. Only the single year 2026 for three arXiv papers is known, which is too thin to score.
Reproducibility55/100
Scored 55/100 as the mean reproducibility readiness over one reviewed paper (reproPapers = 1). That paper — the latent-steering reasoning work — is rated medium readiness: it is training-free and reports consistent Pass@4 gains on Qwen3 1.7B to 14B across 9 benchmarks, but Evolve's verdict flags thin statistical rigor, so reported gains deserve caution until independently replicated.
Research summary
Ruobing Xie is a researcher at Tencent (China) whose cataloged output centers on large language model reasoning. The Evolve catalog lists three 2026 arXiv papers, all with middle authorship and none as corresponding author. The first introduces a training-free, state-conditioned latent steering method with progress guidance: Evolve's review reports consistent Pass@4 gains on Qwen3 models from 1.7B to 14B across 9 benchmarks, while flagging thin statistical rigor and rating reproducibility readiness as medium. The second, Higher-Dimensional Rotary Position Embedding, addresses the architectural side of positional encoding; the third, Dynamic Important Example Mining for Reinforcement Finetuning, targets example selection for RFT training. The bibliometric snapshot records 27 distinct coauthors, 12 of whom recur, and a field distribution confined to a single area — natural language processing (LLM reasoning) — yielding a maximum focus score of 100. Collaboration scores 66 and reproducibility averages 55 over one reviewed paper. Output, impact, and activity metrics are not yet available. The profile is name-resolved from public bibliographic data.
Collaboration pattern
The snapshot records 27 distinct coauthors, with 12 recurring collaborators — Samm Sun, Haoru Tan, Sitong Wu, Yanfeng Chen, Shizhen Zhao, Yang-Tian Sun and others — pointing to a stable core team within a moderately wide network (collaboration score 66/100). On all three 2026 cataloged papers Xie appears as a middle author and not the corresponding author, consistent with contributing inside a team-led research setting at Tencent.
Research keywords
Highlights
Training-free latent steering for efficient LLM reasoning (SPS)
Higher-Dimensional Rotary Position Embedding (2026)
A 2026 arXiv preprint cataloged in Evolve; the title indicates work extending rotary position embeddings to higher dimensions. It has not yet been reviewed by Evolve, so no verdict, keywords, or reproducibility rating is available.
Dynamic Important Example Mining for Reinforcement Finetuning (2026)
A 2026 arXiv preprint cataloged in Evolve; the title indicates work on dynamically mining important training examples for reinforcement finetuning. No Evolve verdict or reproducibility rating exists yet for this paper.
Timeline
- 2026Co-authors 'Efficient Reasoning Exploration via State-Conditioned Latent Steering with Progress Guidance' on arXiv (middle author); Evolve's review notes consistent Pass@4 gains on Qwen3 1.7B–14B across 9 benchmarks.
- 2026Co-authors 'Higher-Dimensional Rotary Position Embedding' on arXiv; the paper is cataloged by Evolve but not yet reviewed.
- 2026Co-authors 'Dynamic Important Example Mining for Reinforcement Finetuning' on arXiv; the paper is cataloged by Evolve without a verdict yet.
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 scores (collaboration, focus, reproducibility) are computed from OpenAlex-derived bibliometric data only; no other sources are included.
- Output, impact, and activity axes are unscored because the snapshot contains no publication-count or citation metrics.
- No ORCID or country field is provided; the profile is name-resolved from public bibliographic data, so identity should be verified before reuse.
- Reproducibility rests on a single reviewed paper (reproPapers = 1), and only that paper carries an Evolve verdict; the other two catalog entries are unreviewed.