Chen Yang
The Hong Kong University of Science and Technology (Guangzhou) · China
Chen Yang — statistical inference for LLM evaluation and energy-measured LLM serving · HKUST (Guangzhou)
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 axis measures publication volume and productivity over time, but the underlying record is insufficient here: OpenAlex links only two 2026 arXiv works, with h-index and i10-index both recorded as 0. No reliable output score is computed.
Impact—
The impact axis would measure citations and scholarly influence, yet no usable citation record is available: the profile shows h-index 0 and i10-index 0, and both 2026 works are too recent to have accrued citations. No impact score is estimated.
Focus0/100
The computed focus score is 0, and the raw field distribution explains why: it splits exactly 50/50 between statistics for machine-learning evaluation (LLM reasoning benchmarks) and energy-efficient AI systems / LLM serving. Topic counts (Topic Modeling ×2, plus single counts for reinforcement learning in robotics, machine learning and algorithms, green IT, and energy efficiency) confirm breadth across evaluation methodology and systems efficiency rather than one narrow specialty.
Collaboration44/100
The collaboration score is 44, mid-range on the 0-100 scale. The record lists 8 distinct co-authors, with frequent collaborators including Xianyang Zhang, Jun Chen, Muhammad Abdur Rab Siddiqui, Daniela Rojas, Wenqi Cui, Yuanyuan Shi, and Yize Chen. Both catalog papers are multi-author efforts: the energy-routing study places Chen Yang in a middle authorship position, while the RLVR inference paper is first-authored but not corresponding.
Activity32/100
The activity score is 32, on the low side, reflecting a very recent record: countsByYear shows only 2 works in 2026 and recent3yWorks equals 2, matching the two arXiv papers in the Evolve catalog. With h-index 0, the recorded activity is a starting point rather than an established pattern; both papers appearing in the same year suggests output is concentrated in 2026.
Reproducibility78/100
Reproducibility is the strongest recorded axis at 78, computed across 2 repro-assessed papers (reproMean 78). The first-authored RLVR pass@k paper rates high readiness, backed by a pre-registered paired-inference framework with simultaneous confidence bands; the energy-routing paper rates medium readiness. Statistical rigor on the evaluation side and measurement-based but harder-to-replicate GPU energy experiments on the systems side fit this score.
Research summary
Chen Yang is affiliated with The Hong Kong University of Science and Technology (Guangzhou) and is profiled here from OpenAlex bibliographic data together with two 2026 arXiv papers reviewed by Evolve. The recorded profile splits evenly — 50/50 in field distribution — between statistics for machine-learning evaluation (LLM reasoning benchmarks) and energy-efficient AI systems / LLM serving. The first-authored paper, 'RLVR is a Kernel, Not a Function: Statistical Inference for pass@k Crossovers,' builds a pre-registered paired-inference framework using Markov-kernel formulations, simultaneous confidence bands, and a multiplier bootstrap to test pass@k crossovers; Evolve finds it rigorous, with only one small, token-limit-dependent reversal statistically confirmed, and rates its reproducibility readiness high. The middle-authored 'Measured Joules, Learned Routes: Learning to Route for Energy-Efficient LLM Serving' trains routers with KL-GRPO on measured GPU joules, reaching 60.9% accuracy at 1692 J, while noting that a fixed Qwen-7B arm (64.0%, 282 J) beats all learned routers on both metrics — an honest negative result. Topic counts (Topic Modeling ×2; reinforcement learning in robotics, machine learning and algorithms, green IT, energy efficiency ×1 each) and 8 distinct co-authors round out the picture. The reproducibility axis scores 78 across two assessed papers, the strongest recorded dimension; output and impact axes lack sufficient citation data.
Collaboration pattern
The record shows 8 distinct co-authors and a collaboration score of 44. Frequent collaborators include Xianyang Zhang, Jun Chen, Muhammad Abdur Rab Siddiqui, Daniela Rojas, Wenqi Cui, Yuanyuan Shi, and Yize Chen. Authorship roles differ by project: first author (non-corresponding) on the RLVR evaluation-statistics paper and a middle position on the energy-routing systems paper, suggesting a pattern of contributing statistical expertise within larger applied teams.
Research keywords
Highlights
Pre-registered statistical inference for pass@k crossovers
Measured-joule LLM routing with an honest negative result
Dual-domain profile: evaluation statistics meets energy-aware serving
The OpenAlex record splits 50/50 between statistics for machine-learning evaluation and energy-efficient AI systems / LLM serving, with a reproducibility mean of 78 across two assessed papers — the strongest recorded dimension. Output and impact axes remain unscored for lack of citation data.
Timeline
- 2026First-authored 'RLVR is a Kernel, Not a Function: Statistical Inference for pass@k Crossovers' appears on arXiv; Evolve rates its reproducibility readiness high.
- 2026Middle-authored 'Measured Joules, Learned Routes: Learning to Route for Energy-Efficient LLM Serving' appears on arXiv; Evolve rates its reproducibility readiness medium.
- 2026OpenAlex records 2 works in 2026 (recent3yWorks = 2) with 8 distinct co-authors — an early-stage record concentrated in a single 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 OpenAlex only; no ORCID is linked, and 'Chen Yang' is a common name, so identity resolution may be imperfect.
- Output and impact dimensions are unscored because the record is very recent (2 works in 2026) with h-index and i10-index recorded as 0.
- Deep analysis covers only the two 2026 arXiv papers reviewed by Evolve; other works may exist outside this catalog.
- The frequent-coauthor list is drawn from OpenAlex co-authorship records and may include name ambiguities.