NLP researcher at University of Michigan focused on LLM reasoning evaluation and dataset representation analysis
2
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—
Output is not scored: this snapshot contains no aggregate publication counts for this researcher. The axis would measure the volume and consistency of papers over time; only two 2026 catalog entries are currently visible, so no estimate is made.
Impact—
Impact is not scored: no citation counts or citation-based indices are available in this snapshot. The axis would measure how widely the two 2026 papers and any prior work have been cited and adopted by other researchers.
Focus100/100
Focus scores 100 because the field distribution contains exactly one assigned field: natural language processing (LLM reasoning evaluation). Both 2026 catalog entries sit inside NLP, covering step-level process reward modeling for mathematical reasoning and country-aware dataset representation, so the visible work is tightly concentrated. With only two classified papers, the perfect score may partly reflect catalog coverage rather than long-term specialization.
Collaboration62/100
Scoring 62, collaboration rests on 22 distinct coauthors in the raw bibliometric data and a twelve-person recurring co-author group, including Yiming Feng, Yulong Chen, Rada Mihalcea, and Joan Nwatu. Both cataloged 2026 papers list Deng as a middle, non-corresponding author, consistent with contributor roles inside larger teams rather than sole-leadership projects. The breadth of the co-author pool supports the above-median score, though position evidence spans only two entries.
Activity—
Activity is not scored: recent-year publishing cadence cannot be computed from this snapshot, which only shows papers dated 2026. The axis would measure ongoing research throughput across recent years, and it needs multi-year publication data before any assessment is possible.
Reproducibility55/100
Reproducibility scores 55 as the mean readiness across one assessed paper (reproPapers = 1). The scored entry, LLaDA-PRM, is rated medium: Evolve's review credits a controlled 54-run ablation as methodological support for bidirectional attention in step-level PRMs, while flagging that model scale and the attention mechanism are conflated. A single-paper base means the score should be read as an early signal rather than a stable estimate.
Research summary
Naihao Deng is a researcher at the University of Michigan working in natural language processing, with a cataloged focus on large language model reasoning evaluation. The Evolve catalog lists two 2026 arXiv papers: LLaDA-PRM, a bidirectional step-level reasoning evaluator built on the LLaDA diffusion language model, and AtlasNLP, a country-aware atlas of dataset representation in NLP. In LLaDA-PRM, a controlled 54-run ablation supports bidirectional attention for step-level process reward models, and the 8B model sets state of the art on an MR benchmark, though Evolve's review notes that the effects of model scale and attention remain conflated. The radar profile shows maximal focus (100), derived from a single assigned field, natural language processing (LLM reasoning evaluation); a collaboration score of 62, grounded in 22 distinct coauthors and a twelve-person recurring co-author group; and a reproducibility score of 55 from one assessed paper at medium readiness. Deng appears as a middle, non-corresponding author on both cataloged papers, consistent with contributor roles inside larger teams. Output, impact, and activity axes are not scored because the underlying bibliometric data is not yet available. This is a full but early-stage profile built from a small, recent slice of work.
Collaboration pattern
Naihao Deng's cataloged work is team-based: 22 distinct coauthors appear in the raw data, anchored by a twelve-person recurring group including Yiming Feng, Yulong Chen, Rada Mihalcea, Joan Nwatu, Tsedeniya Solomon Amare, Longju Bai, Bontu Fufa Balcha, Zayd Bashir, Angana Borah, Zara Burzo, Yubin Choi, and Samika Gupta. On both 2026 papers Deng is a middle, non-corresponding author, indicating contributor roles within larger NLP teams rather than first- or sole-author leadership.
Research keywords
Highlights
Bidirectional attention for step-level process reward models
8B model sets MR-benchmark SOTA, with a conflation caveat
Country-aware atlas of dataset representation in NLP
AtlasNLP maps how datasets are represented across countries in NLP research; it is cataloged as a 2026 arXiv paper and has not yet received an Evolve review verdict or reproducibility assessment.
Timeline
- 2026First cataloged entry: LLaDA-PRM, a bidirectional step-level reasoning evaluator, appears on arXiv with Deng listed as a middle, non-corresponding author.
- 2026Second cataloged entry: AtlasNLP, a country-aware atlas of dataset representation in NLP, appears on arXiv, again with Deng as a middle, non-corresponding author.
- 2026Evolve computes the radar snapshot for this researcher: focus 100, collaboration 62, reproducibility 55 from one assessed paper, with 22 distinct coauthors in the raw data.
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 coverage is sparse: the output, impact, and activity axes lack underlying data, and no citation counts are available in this snapshot.
- The reproducibility score (55) averages a single assessed paper, so it rests on a very small evidence base.
- The profile is name-resolved from public bibliographic data; no ORCID was provided, so identity confidence is limited for this name.
- Both cataloged papers are 2026 arXiv preprints; peer-reviewed venue status is not reflected in the provided data.