Shenzhen Research Institute of Big Data researcher: LLM evaluation, graph learning, speech models, medical imaging
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 scored yet: the current snapshot contains no indexed publication-volume data for this researcher beyond the four cataloged arXiv papers. The output axis would measure the count, venue quality, and year-over-year consistency of publications.
Impact—
Not scored yet: citation counts, h-index, and per-paper influence are absent from the input data, so community uptake cannot be assessed. The impact axis would measure how strongly the field cites and builds on this researcher's work.
Focus0/100
The radar focus score is 0, and the raw field distribution explains it: the four cataloged 2026 papers each hold exactly 0.25 share across LLM evaluation methodology, speech and NLP, medical image analysis, and graph representation learning. This is an evenly spread, exploratory portfolio rather than a concentrated agenda; more papers are needed to see whether a dominant direction emerges.
Collaboration70/100
Collaboration scores 70, backed by a raw count of 32 distinct coauthors. All four cataloged papers are multi-author works: middle position on three, last position on the mammography paper, and none flagged as corresponding author. Frequent coauthors include Qianli Shen, Ruomeng Ding, Yanxi Chen, Daoyuan Chen, Yaliang Li, Junyi Ao, Kainan Peng, Mingbo Ma, Shun Zhang, and Zhenyu Tang.
Activity—
Not scored yet: the snapshot contains only four 2026 papers with no full publication history, so cadence and recency trends cannot be computed. The activity axis would measure publication frequency and recent-year productivity over time.
Reproducibility69/100
Reproducibility scores 69, the mean readiness over 4 scored papers (reproMean 69, reproPapers 4). The LLM-evaluation and QUARTET papers rate high; the mammography paper rates medium given single-seed runs, tiny data, and missing ablations; the spoken-language paper rates low owing to proprietary data and a missing interleaving baseline. The profile leans toward verifiable methods.
Research summary
Xiang Li is a researcher at the Shenzhen Research Institute of Big Data whose cataloged work spans four areas in equal measure — LLM evaluation methodology, speech and natural language processing, medical image analysis, and graph representation learning — with the raw field distribution assigning each 0.25. The Evolve catalog holds four 2026 arXiv papers. Three are methods contributions with quantified gains: hierarchical-Bayes Neyman allocation with asynchronous speculation that cuts stochastic LLM evaluation variance 13–34% versus uniform sampling; QUARTET, a quad-branch cross-attention graph transformer with causal personalized-PageRank sampling that beats RelGT on 8 of 12 RelBench tasks; and an asynchronous think-aloud decoding scheme for spoken language models that shrinks reasoning silence from 12.8s to 0.36s at near chain-of-thought accuracy. The fourth, M3D-Net, reports 97.78% mammography validation accuracy but rests on single-seed runs, tiny data, and no ablations. Across the four papers, reproducibility readiness averages 69 (high, high, medium, low), and the collaboration axis scores 70 with 32 distinct coauthors. Output, impact, and activity scores are not yet available. Xiang Li is a common name; this profile is name-resolved from public bibliographic data.
Collaboration pattern
All four cataloged papers are multi-author works, with Xiang Li in the middle position on three and the last position on the mammography paper; none is flagged as corresponding author. The raw data counts 32 distinct coauthors (collaboration score 70). Recurring names include Qianli Shen, Ruomeng Ding, Yanxi Chen, Daoyuan Chen, Yaliang Li, Junyi Ao, Kainan Peng, Mingbo Ma, Shun Zhang, and Zhenyu Tang; the list also repeats 'Xiang Li' itself, a name-disambiguation artifact.
Research keywords
Highlights
Variance-reduced LLM evaluation via Neyman allocation
QUARTET pushes relational deep learning past RelGT
M3D-Net reports 97.78% mammography validation accuracy
Thinking aloud while speaking: reasoning silence from 12.8s to 0.36s
Timeline
- 2026Co-authored 'Spoken Language Models that Think Aloud', an asynchronous think-aloud decoding scheme that cuts SLM reasoning silence from 12.8s to 0.36s at near chain-of-thought accuracy.
- 2026Co-authored 'QUARTET', a quad-branch cross-attention graph transformer with causal personalized-PageRank sampling that beats RelGT on 8 of 12 RelBench tasks.
- 2026Last-position co-author on 'M3D-Net', a mammography classification encoder reporting 97.78% validation accuracy from coordinate attention, differential attention, and dynamic dense feature reuse.
- 2026Co-authored 'Speculative Evaluation of Stochastic LLMs', applying hierarchical-Bayes Neyman allocation with asynchronous speculation to cut evaluation variance 13–34% versus uniform sampling.
Papers in the Evolve catalog
Speculative Evaluation of Stochastic LLMs
M3D-Net: Hierarchical Coordination of Spatial Context, Feature Reuse, and Differential Attention for Mammography Classification
QUARTET: Quad-branch cross-Attention and Random-walk Traces for Enhancing Transformers on Relational Graphs
Spoken Language Models that Think Aloud
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 are computed from public bibliometric data (OpenAlex-based); coverage and disambiguation may lag or mis-resolve common names.
- 'Xiang Li' is a common name; this profile is name-resolved from public bibliographic records, and the raw coauthor list contains the researcher's own name as an artifact.
- All cataloged papers are 2026 arXiv preprints; the data includes no peer-review venue, no citation counts, and no ORCID identifier.
- Output, impact, and activity axes lack underlying data and are reported as missing rather than estimated.