Jianfeng Xu

—

Unclaimed · AI-structured profileOpenAlex

ML researcher focused on knowledge editing for LLMs, VLM evaluation methodology, and temporal KG forecasting

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-28) 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: this OpenAlex identity shows h-index 0 and i10-index 0, so the output axis lacks sufficient data. It would measure total publication volume, authorship positions, and year-by-year productivity.

Impact—

Not scored: with h-index and i10-index both at 0 in OpenAlex, citation impact cannot be assessed for this identity. This axis would measure citation counts, citing works, and field-normalized influence of the reviewed papers.

Focus0/100

This identity scores 0 on focus, reflecting a split topical signature: the field distribution divides evenly (0.5 knowledge-editing NLP, 0.5 machine learning), while OpenAlex topics attach most records to wireless communication security (3) and modulation classification (2), areas absent from the three reviewed ML/NLP papers. The reviewed work itself is tightly themed around LLM knowledge editing and evaluation, so the zero likely reflects identity fragmentation rather than genuine scattering.

Collaboration44/100

Collaboration scores 44. The identity lists 8 distinct co-authors in the recent window, with a recurring core team — Zeyan Li, Hu Xu, Siyuan Qiu, Libing Chen, Shengda Zhuo, and Yin Tang — appearing across the catalog. On all three reviewed papers Jianfeng Xu takes the last author position without corresponding status, a senior-contributor pattern within a compact, stable group rather than a large distributed network.

Activity47/100

Activity scores 47. OpenAlex records 4 works in the recent three-year window, all of them landing in 2026 (countsByYear shows 2026: 4 only), and all three catalog papers are 2026 arXiv preprints. This is a concentrated, recent burst of output rather than a sustained multi-year cadence, which keeps the activity score moderate despite strong current momentum.

Reproducibility55/100

Reproducibility scores 55, computed as a mean over 2 papers. All three reviewed papers carry medium reproducibility readiness: ALOE and BridgeMem report concrete numbers (eff .955-.999, loc .981-1; 20/20 metric wins across 5 benchmarks), but Evolve's verdicts flag undertuned baselines, missing variance, and absent significance testing as limits to independent replication.

Research summary

Jianfeng Xu is a machine learning researcher whose recent work centers on how large models store, retrieve, and report knowledge. The Evolve catalog holds three 2026 arXiv papers, on all of which Xu is the last author without corresponding status: ALOE (semantically addressed gated low-rank operators for knowledge editing, with reported edit efficacy of .955-.999 and locality of .981-1), a study showing that chain-of-thought prefix scoring understates vision-language model accuracy by 35.28 points on ScienceQA, and BridgeMem (a causal dyadic-transition residual method winning all 20 metric comparisons across 5 temporal knowledge graph forecasting benchmarks). Together they sketch a coherent agenda: modifying what LLMs know, measuring what VLMs actually answer, and forecasting link evolution in temporal knowledge graphs. The OpenAlex identity behind this profile scores 47 for activity (4 works in the recent three-year window, all in 2026), 44 for collaboration (8 distinct co-authors, with a recurring core group), and 55 for reproducibility (mean over 2 papers). Output and impact are not scored: h-index and i10-index are recorded as 0. Notably, the identity's OpenAlex topics are dominated by wireless communication security (3) and modulation classification (2), which do not overlap the reviewed papers, so this profile is name-resolved from public bibliographic data and should be read with that caveat.

Collaboration pattern

Collaboration is compact and stable: the OpenAlex identity lists 8 distinct co-authors, and a recurring core — Zeyan Li, Hu Xu, Siyuan Qiu, Libing Chen, Shengda Zhuo, Yin Tang — recurs across the catalog. On all three reviewed 2026 papers Xu is the last author without corresponding status, consistent with a senior supervisory role in a small team. The moderate collaboration score of 44 reflects this small-cohort pattern rather than broad external networking.

Research keywords

knowledge editinglarge language modelslow-rank updatesgated operatorsvision-language modelschain-of-thoughtmultiple-choice evaluationselection biaslogit analysistemporal knowledge graph forecastinglink predictionresidual scoring

Highlights

ALOE: gated low-rank operators for LLM knowledge editing
The 2026 ALOE paper proposes semantically addressed low-rank updates with gated operators for editing facts in large language models; Evolve's review notes reported edit efficacy of .955-.999 and locality of .981-1, calling it strong mechanism work tempered by undertuned baselines and weak generalization evidence.
CoT-prefix scoring understates VLM accuracy by 35.28 points
In a 2026 evaluation-methodology study, reading label logits after a reasoning request (CoT-prefix scoring) was shown to understate vision-language model accuracy by 35.28 points on ScienceQA; Evolve calls the evidence convincing, though coverage is narrow and unaccompanied by significance testing.
BridgeMem wins all 20 metric comparisons on 5 TKG benchmarks
The 2026 BridgeMem paper adds causal dyadic-transition residuals on top of a frozen CENET backbone for temporal knowledge graph forecasting, winning all 20 metric comparisons across 5 benchmarks; Evolve flags the cheap, controlled design but notes tiny WIKI margins lacking variance estimates.

Timeline

2026OpenAlex records 4 works in the recent three-year window, all dated 2026 — a concentrated first burst of indexed output.
2026Released ALOE, gated semantically addressed low-rank operators for knowledge editing in large language models (reported efficacy .955-.999, locality .981-1).
2026Published the evaluation-methodology study showing CoT-prefix scoring understates VLM accuracy by 35.28 points on ScienceQA.
2026Released BridgeMem, causal dyadic-transition residuals winning all 20 metric comparisons across 5 temporal knowledge graph benchmarks.

Papers in the Evolve catalog

Frequent co-authors

Zeyan Li×2 Hu Xu×1 Siyuan Qiu×1 Zeyan Li×1 Libing Chen×1 Shengda Zhuo×1 Yin Tang×1 Jianfeng Xu×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.

Bibliometrics come from OpenAlex only; h-index and i10-index are recorded as 0, so the output and impact axes carry no score.
The OpenAlex identity's topics are dominated by wireless communication security (3) and modulation classification (2), which do not overlap the reviewed ML/NLP papers — this profile is name-resolved from public bibliographic data and common-name ambiguity is possible.
No institution, country, or ORCID was provided, so affiliation and identity confidence are unverified.
Reproducibility (55) is a mean over only 2 papers; the frequent-coauthor list also contains duplicates and the researcher's own name.