Wei Xiong
University of International Business and Economics
UIBE machine learning researcher: LLM continual learning and streaming supervised dimension reduction
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-14) 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—
Insufficient data: this profile rests on two cataloged arXiv papers, so publication volume, venue distribution and productivity trends over time cannot be scored. The output axis would measure total and per-year publication counts, authorship roles and venue quality.
Impact—
Insufficient data: no citation counts, h-index or downstream-usage signals are available for this researcher yet. The impact axis would measure citations to the cataloged papers, their uptake in follow-up work, and influence on the LLM continual-learning and dimension-reduction communities.
Focus0/100
The focus dimension scores 0, and the raw field distribution explains why: cataloged output splits exactly evenly (0.5/0.5) between two distinct lines — LLM fine-tuning and continual learning, and streaming supervised dimension reduction. The two lines share a machine-learning core (Jacobian null-space analysis and supervised PCA) but address different problem settings, so the profile currently reads as two-track rather than narrowly specialized.
Collaboration50/100
Collaboration scores 50 on a 0-100 scale, with 11 distinct coauthors recorded. The frequent-coauthor list is headed by Zhilong Zheng, Letian Tao and Yang Guan, followed by Yujie Yang, Kehua Sheng, Bo Zhang, Jingliang Duan, Keqiang Li, Shengbo Eben Li and Zhenlin Yao. Authorship positions span middle author on the JANUS paper and last author on the OKSPCA paper, suggesting participation in team-based projects rather than a single-author mode.
Activity—
Insufficient data: with only two 2026 arXiv entries and no multi-year publication record in this snapshot, research cadence cannot be scored. The activity axis would measure publication frequency over time, recency of output, and continuity of research across years.
Reproducibility78/100
Reproducibility scores 78, the mean across the 2 papers Evolve has assessed. The streaming dimension-reduction paper rates high readiness — Evolve calls it rigorous and honest — while the JANUS continual-learning paper rates medium: its first-order theory is reasonable, but testing is narrow and the replay set overlaps the evaluation set. The overall profile suggests methodological care, tempered by evaluation-scope limits on one paper.
Research summary
Wei Xiong is a machine learning researcher at the University of International Business and Economics whose cataloged work splits evenly between two lines: LLM fine-tuning and continual learning, and streaming supervised dimension reduction, each accounting for 0.5 of the field distribution. The Evolve catalog holds two 2026 arXiv papers. On the continual-learning side, Xiong is a middle author on 'Past, Future, All at Once', which proposes post-hoc Jacobian null-space (JANUS) rectification to ease the stability-plasticity dilemma in fine-tuned LLMs; Evolve's review finds the first-order theory reasonable but notes a narrow test scope and overlap between replay and evaluation sets. On the statistical side, Xiong is the last author on 'Online Supervised Dimension Reduction with Random Features', whose OKSPCA diagnostics show that high objective capture can coexist with large subspace error — a study judged rigorous and honest despite tuning asymmetry and unresolved causes. Reproducibility scores 78 as the mean across two assessed papers (one medium, one high readiness); collaboration scores 50 with 11 distinct coauthors; focus scores 0, consistent with the even two-way field split. Output, impact and activity axes lack sufficient data. Given a common name and no ORCID, this profile is name-resolved from public bibliographic data.
Collaboration pattern
With 11 distinct coauthors and a collaboration score of 50, Xiong works in team settings rather than solo. The recurring coauthor list — including Zhilong Zheng, Letian Tao, Yang Guan, Yujie Yang, Kehua Sheng, Bo Zhang, Jingliang Duan, Keqiang Li, Shengbo Eben Li and Zhenlin Yao — spans more names than the two cataloged papers alone would imply, which may reflect name-resolution noise for this common name; coauthor overlap across the two field lines is not yet established.
Research keywords
Highlights
Post-hoc JANUS rectification targets forgetting in LLM fine-tuning
Diagnostics show high objective capture can hide large subspace error
Timeline
- 2026Co-authored the arXiv paper 'Past, Future, All at Once' in a middle authorship position, proposing post-hoc JANUS Jacobian null-space rectification to mitigate the stability-plasticity dilemma in LLM fine-tuning.
- 2026Co-authored the arXiv paper 'Online Supervised Dimension Reduction with Random Features' as last author, delivering OKSPCA diagnostics and computational trade-off analysis for streaming supervised PCA.
- 2026Both cataloged papers appeared in the same year, splitting evenly between LLM continual learning and streaming supervised dimension reduction — the basis of this early-stage two-track profile.
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.
- Dimension scores are computed from the provided bibliometric snapshot only; the output, impact and activity axes lack sufficient data and are not scored.
- 'Wei Xiong' is a common name with no ORCID provided; the profile is name-resolved from public bibliographic data and the frequent-coauthor list may contain disambiguation noise.
- Only two 2026 arXiv preprints are in the Evolve catalog; peer-review status, citation counts and publication history beyond this snapshot are unknown.
- Country and ORCID fields are missing in the source data; the institution is reported as provided.