Yulong Chen

University of Aberdeen

Unclaimed · AI-structured profileOpenAlex

University of Aberdeen NLP researcher: bidirectional process reward models and fairness-benchmark stress tests.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

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—

Not yet scored: the data contains no publication count, venue list, or yearly output totals for this researcher. The output axis would measure total and per-year paper volume once a fuller bibliographic record is linked.

Impact—

Not yet scored: no citation counts, h-index, or downstream uptake measures are available in the source data. The impact axis would quantify citations and influence of the catalogued work once citation records are attached.

Focus0/100

Focus scores 0 in the current data: the two catalogued 2026 papers split evenly, at 0.5 each, between LLM reasoning evaluation and LLM alignment and fairness evaluation, so no single subfield dominates yet. Both works nonetheless share the theme of LLM evaluation methodology — step-level reward modeling and benchmark stress-testing — hinting at a coherent direction, but a two-paper sample is too small to confirm a stable focus.

Collaboration41/100

Collaboration scores 41. The two catalogued papers involve 7 distinct coauthors, a modest-size network for a full profile. Recurring names in the co-occurrence record are Naihao Deng and Rada Mihalcea (each listed twice), alongside Yiming Feng, Samee Arif, and Shuaichen Chang. Chen holds a middle author position on both papers and is corresponding author on neither, consistent with contributing inside larger teams rather than leading them so far.

Activity—

Not yet scored: apart from the two 2026 arXiv papers, no multi-year publication series exists in the data, so research cadence cannot be measured. The activity axis would track per-year output and publishing gaps over time.

Reproducibility78/100

Reproducibility scores 78, the mean across 2 reviewed papers (reproMean 78, reproPapers 2). The fairness study carries high reproducibility readiness, while LLaDA-PRM is rated medium: its 54-run controlled ablation documents design choices carefully, but Evolve notes the headline result conflates model scale with attention type, which complicates independent replication of the 8B model's benchmark state of the art.

Research summary

Yulong Chen is an NLP researcher at the University of Aberdeen whose catalogued work centers on evaluating large language models along two axes: how well they reason step by step, and how fairly they behave after alignment. The Evolve catalog contains two 2026 arXiv papers, on both of which Chen appears in a middle author position. The first, LLaDA-PRM, builds a bidirectional step-level process reward model on the LLaDA diffusion language model; Evolve's review credits a controlled 54-run ablation supporting the bidirectional-attention design and an 8B model that sets state of the art on an MR benchmark, while flagging that model scale and attention type are conflated. The second shows that BBQ-style fairness benchmarks can be passed from a single example through a shallow abstention cue, with convincing multi-family evidence but a non-transfer claim resting on one probe. Bibliometric scores reflect this early, concentrated record: reproducibility is comparatively strong at 78 (the mean over 2 reviewed papers, with high readiness for the fairness study and medium for the PRM work), collaboration scores 41 with 7 distinct coauthors, and focus scores 0 because the record splits evenly — 0.5 and 0.5 — across LLM reasoning evaluation and LLM alignment and fairness evaluation. Output, impact, and activity metrics are not yet available. Recurring coauthors include Yiming Feng, Naihao Deng, Rada Mihalcea, Samee Arif, and Shuaichen Chang.

Collaboration pattern

Chen's catalogued papers involve 7 distinct coauthors, and the collaboration axis scores 41. The recurring names are Naihao Deng and Rada Mihalcea (each listed twice in the co-occurrence record), alongside Yiming Feng, Samee Arif, and Shuaichen Chang. On both papers Chen sits in a middle author position rather than first or corresponding, suggesting contribution within larger teams. The network is modest in size, consistent with an early-stage record built around two 2026 arXiv preprints.

Research keywords

process reward modelsbidirectional attentiondiffusion language modelsLLaDAstep-level evaluationmathematical reasoningfairness evaluationBBQ benchmarkGRPOin-context learningRLHFsocial bias

Highlights

Bidirectional process reward model reaches MR-benchmark state of the art
LLaDA-PRM adapts the LLaDA diffusion language model with bidirectional attention for step-level process reward modeling. Evolve's review credits a controlled 54-run ablation supporting the bidirectional-attention design and an 8B model that sets state of the art on the MR benchmark, while cautioning that model scale versus attention type remains conflated.
One example can saturate BBQ-style fairness benchmarks
This study on aligned LLMs shows that BBQ-style fairness benchmarks can be passed from a single example via a shallow abstention cue. Evolve finds the multi-family evidence convincing but notes that the non-transfer claim rests on a single probe.
Two 2026 papers, both evaluation-focused, with reproducibility mean of 78
Both catalogued papers target LLM evaluation methodology. The fairness study carries high reproducibility readiness and the PRM work medium, yielding a reproducibility mean of 78 across the 2 reviewed papers — the strongest quantitative signal in this profile.

Timeline

2026Co-authored 'LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator' on arXiv, in a middle author position, in the field of LLM reasoning evaluation.
2026Co-authored 'One Example Is Enough to Pass Fairness Benchmarks: Rethinking Fairness Evaluation for Aligned LLMs' on arXiv, also in a middle author position.
2026Both papers enter the Evolve review catalog; the reviewed reproducibility mean stands at 78 across 2 papers, with 7 distinct coauthors recorded.

Papers in the Evolve catalog

Frequent co-authors

Yiming Feng×1 Naihao Deng×1 Rada Mihalcea×1 Naihao Deng×1 Samee Arif×1 Shuaichen Chang×1 Rada Mihalcea×1

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 dimension scores are computed from a small sample: 2 reviewed papers and 7 distinct coauthors, so scores may shift substantially as the record grows.
No ORCID or country data was available; 'Yulong Chen' is a common name, so this profile is name-resolved from public bibliographic data and not independently identity-verified.
Output, impact, and activity metrics (publication counts, citations, yearly cadence) are missing from the source data and are not estimated here.
Both catalogued papers are 2026 arXiv preprints; peer-review status and venue acceptance are not captured in the data.