Evolve.Pub
Sign inScan your paper
XL

Xiaoyu Li

UNSW Sydney · China

Unclaimed · AI-structured profileOpenAlex

Early-career UNSW Sydney ML researcher working across NAS, learning theory, interpretability and AI security

6

papers here

6

reviewed by Evolve

1

first-author

0

corresponding

—

works · OpenAlex

0

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: insufficient dataOutputImpact: insufficient dataImpactFocus: 3/100FocusCollaboration: 59/100CollaborationActivity: 84/100ActivityReproducibility: 43/100Reproducibility

Output—

No output score yet: the OpenAlex record lists h-index 0 and i10-index 0 with citation-linked coverage absent, so publication-volume scoring is not currently supported; this axis would measure the scale and consistency of the researcher's publication record.

Impact—

No impact score yet: h-index and i10-index are recorded as 0 and all six cataloged works are 2026 arXiv preprints, too recent for citation accumulation; this axis would measure citation-based influence and scholarly uptake of the work.

Focus3/100

The focus score is very low (3/100), consistent with the raw field distribution: the leading field, machine learning (neural architecture search), accounts for only about one third (0.33) of fielded works, with mechanistic interpretability, statistical learning theory, ML for healthcare and AI security (diffusion watermarking) each around 0.17. Five named OpenAlex topics similarly split across topic modeling, stochastic gradient optimization, RL in robotics, and semigroup and automata theory.

Collaboration59/100

The collaboration score is 59/100, backed by 18 distinct co-authors in the recent window and 12 recurring collaborators, including Jiaojiao Jiang, Zheng Gao, Yifan Yang and José Miguel Hernández-Lobato. Authorship patterns in the six cataloged papers skew middle-position, with one first-author theory paper and no corresponding-author role, suggesting a contributor profile within larger teams rather than a lone or leading PI position.

Activity84/100

The activity score is high (84/100): the snapshot records 17 works in the recent three-year window, all landing in 2026, and Evolve's catalog alone contains six arXiv preprints from that year across NAS, learning theory, interpretability, health AI and AI security. This is a concentrated burst of output within a single year, though the record holds no earlier-year counts to establish a longer trajectory.

Reproducibility43/100

Reproducibility readiness averages 43/100 across the six reviewed papers: RiPPLE, CoRA-NAS, the Rademacher bounds paper and DRIFT rate medium, while SIFPBPNet and the feature-superposition survey rate low — none reaches high readiness. Recurring Evolve concerns include unstated code release, absence of significance tests, and single-backbone evaluation, which currently cap how easily third parties could rerun or verify this work.

Research summary

Xiaoyu Li is an early-career machine learning researcher at UNSW Sydney whose 2026 output spans five subfields: neural architecture search (NAS), statistical learning theory, mechanistic interpretability, machine learning for healthcare, and AI security around diffusion-model watermarking. The OpenAlex snapshot records 17 works in the recent three-year window — all dated 2026 — with 18 distinct co-authors, but an h-index and i10-index of 0, so citation impact cannot yet be assessed. In Evolve's six-paper reviewed catalog, Li is first author of 'Nearly Tight Rademacher Bounds for Sparsely Activated Neural Networks', which removes a √n factor from prior sparsity-aware generalization bounds up to a log^{3/2}(2m) term with matching lower bounds. Elsewhere Li appears as a middle author: the NAS performance predictors RiPPLE and CoRA-NAS, the SIFPBPNet cuffless blood-pressure network (SBP MAE 8.57 mmHg), the DRIFT watermark-removal attack (98-100% success across nine schemes), and a feature-superposition survey. Radar scores show high activity (84) but very low focus (3), and reproducibility readiness averages 43/100, so the profile is promising but early-stage.

Collaboration pattern

Li works in team settings: 18 distinct co-authors in the recent window and 12 recurring collaborators, with Jiaojiao Jiang, Zheng Gao, Yifan Yang, Zhaoyan Wang, Andi Han, Dai Shi and José Miguel Hernández-Lobato among the most frequent. Five of six cataloged papers list Li as a middle author and none as corresponding author, with one first-author theory paper — a contributor role within multi-author groups spanning Sydney and international collaborators.

Research keywords

neural architecture searchzero-cost proxiesperformance predictionlearning curve extrapolationgeneralization boundsRademacher complexityactivation sparsitymechanistic interpretabilitysuperpositioncuffless blood pressure estimationdiffusion model watermarkingstatistical learning theory

Highlights

Timeline

  1. 2026Six reviewed arXiv preprints dated 2026 enter the Evolve catalog, spanning NAS, statistical learning theory, mechanistic interpretability, biomedical signal processing and AI security.
  2. 2026First-author paper 'Nearly Tight Rademacher Bounds for Sparsely Activated Neural Networks' removes a √n factor from prior generalization bounds with matching lower bounds.
  3. 2026Co-authors two NAS performance-prediction papers, RiPPLE and CoRA-NAS, both built on zero-cost proxies and learning curve extrapolation.
  4. 2026Applied contributions appear: the SIFPBPNet wearable cuffless blood-pressure network and the DRIFT diffusion watermark-removal attack.

Papers in the Evolve catalog

Frequent co-authors

Jiaojiao Jiang×4Zheng Gao×3Yifan Yang×2Zhaoyan Wang×2Andi Han×2Dai Shi×1José Miguel Hernández-Lobato×1Zhizhou Sha×1Junbin Gao×1Shuailong Tang×1Donglin Xie×1Wei Chen×1

Is this you? Claim this profile

Enter the email you published with. We'll send a one-time claim link; after claiming you can edit the introduction or hide this page. Requests from a mismatched address go to manual review.

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 figures come from an OpenAlex snapshot only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and 'Xiaoyu Li' is a common name.
  • h-index and i10-index are recorded as 0 with no citation counts available, so the output and impact axes are unscored rather than estimated.
  • All six cataloged works are 2026 arXiv preprints; the data contain no peer-reviewed venue, award or career-history information.
  • Reproducibility ratings and verdicts reflect Evolve's own paper reviews; no earlier-year publication counts exist to establish a multi-year trend.
Xiaoyu Li — Researcher Profile | Evolve.Pub