NUS researcher across LLM post-training, LLM evaluation, and learned physics simulators; strong reproducibility signal
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 behind this profile contains no publication-volume figures beyond the three reviewed 2026 preprints. This axis would measure total output — paper count, venue spread, and first/corresponding-author share — once a fuller bibliographic record is linked.
Impact—
No citation counts or influence signals are available in the input data, so impact is unscored. The axis would track citations to, and downstream uptake of, the DF-OPD, FLY-EVAL++, and TRACE papers once citation sources are linked.
Focus0/100
Scores 0 on focus, and the raw field distribution explains why: output splits almost exactly into thirds — LLM post-training, LLM evaluation and benchmarking, and scientific machine learning for learned physics simulation (0.33 each). Three 2026 papers, three distinct fields, no dominant area yet. At this stage this reads as deliberate breadth — distillation methods, evaluation protocols, GNN simulators — rather than drift, but concentration remains to be established.
Collaboration62/100
Scores 62 on collaboration, built from 22 distinct coauthors across the three reviewed 2026 papers. The pattern looks team-based rather than solitary: papers in all three fields carry mid-list authorship with a recurring collaborator pool — Gengsheng Li, Mao Zheng, Mingyang Song, Jie Sun, and others — suggesting stable small groups per topic rather than one-off partnerships. None of the three reviewed papers lists Fang as corresponding author, so independence on the author line is not yet demonstrated.
Activity—
Recent-activity data — publication cadence, submission rhythm, year-over-year trend — is not part of the provided record, so this axis is unscored. It would measure how active the researcher is over time; the three papers' shared 2026 dates are the only hint.
Reproducibility70/100
Scores 70, the mean repro-readiness across the three reviewed papers. The spread is informative: FLY-EVAL++ rates high thanks to deterministic verifiers and an evidence-driven protocol, while DF-OPD and TRACE sit at medium — Evolve flags that DF-OPD's 0.13-point headline gaps lack error bars and that TRACE's near-zero overlap is enforced rather than learned. Reproducibility looks like a design goal, strongest where evaluation is deterministic.
Research summary
Junfeng Fang is a researcher affiliated with the National University of Singapore whose reviewed work splits almost evenly across three fields: LLM post-training, LLM evaluation and benchmarking, and scientific machine learning for learned physics simulation. The Evolve catalog holds three 2026 arXiv papers, each with Fang in a middle author position. Data-free On-policy Distillation introduces DF-OPD, arguing that on-policy distillation barely needs its training data because a teacher can write its own prompts; Evolve calls the study provocative and well-controlled but notes the 0.13-point headline gaps lack error bars. FLY-EVAL++ contributes a deterministic, evidence-driven evaluation protocol for safety-constrained LLM flight prediction, rated high for reproducibility. TRACE proposes edge-level contact memory for GNN-based granular simulators, beating GNS and NMGNS on sand column collapse. Three signals stand out: a reproducibility score of 70 (mean over three papers), a collaboration score of 62 built on 22 distinct coauthors, and a focus score of 0 reflecting the even three-way field split. Citation, output, and activity metrics are not yet available, so those axes remain unscored in this early-stage profile.
Collaboration pattern
Collaboration is team-based: the three 2026 papers share a pool of 22 distinct coauthors, with frequent collaborators including Gengsheng Li, Mao Zheng, Mingyang Song, Jie Sun, Zeyuan Liu, and Ruiqi Liu. Fang appears as a middle-position author on all three reviewed papers and as corresponding author on none, consistent with contributing inside established small groups across three fields rather than leading them — the 62 collaboration score reflects breadth of contacts, not yet demonstrated independent leadership.
Research keywords
Highlights
Data-free on-policy distillation: teachers can write their own prompts
A deterministic, evidence-driven eval protocol for LLM flight prediction
Edge-level contact memory for GNN granular simulators
Timeline
- 2026Data-free On-policy Distillation posted to arXiv — middle author, LLM post-training.
- 2026FLY-EVAL++ posted to arXiv — middle author, LLM evaluation and benchmarking.
- 2026TRACE posted to arXiv — middle author, scientific machine learning.
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.
- All bibliometric scores derive from a small public bibliographic snapshot: three reviewed 2026 arXiv papers and 22 distinct coauthors.
- 'Junfeng Fang' is a common name; this profile is name-resolved from public bibliographic data and identity consolidation should be independently verified.
- Citation counts, total output, and recent-activity metrics are absent from the input data, so the impact, output, and activity axes are unscored.
- All reviewed papers are 2026 arXiv preprints; peer-review status and venue acceptance are not reflected.