Peilin Zhao

Shanghai Jiao Tong University

Unclaimed · AI-structured profileOpenAlex

Researcher at Shanghai Jiao Tong University bridging adaptive optimization theory and embodied world models.

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—

No publication-volume metrics are available in the current snapshot, so output is unscored rather than estimated. This axis would measure total paper count, venue distribution, and yearly production; the only dated evidence so far is two 2026 arXiv papers.

Impact—

Citation-based metrics are absent from the input data, so impact is unscored. This axis would measure citations, h-index, and citing works, i.e. the influence of the record within optimization theory and embodied AI.

Focus0/100

Evolve scores focus at 0, the floor of the scale, and the raw field distribution explains why: the catalogued record splits exactly evenly — 50% robotics and embodied AI, 50% machine learning (optimization theory). The two clusters address different communities, problems, and methods — adaptive subgradient theory versus embodied world-model surveys — so no dominant direction has emerged yet. On this evidence it reads as deliberate breadth rather than drift, but focus remains unestablished.

Collaboration58/100

Collaboration scores 58 out of 100, supported by 17 distinct coauthors in the snapshot. The frequent-collaborator list — Nanjie Yao, Hao Wang, Chong Cheng, Zhikang Chen, Wenzhe Li, Jiafei Lyu, Li Shen, Zongqing Lu, Gao Huang, Steven Hoi, Dacheng Tao, Deheng Ye — spans both of his fields. On the two catalogued papers he is last author on one and middle-list on the other, neither as corresponding author, consistent with a contributing-collaborator role.

Activity—

Activity is unscored because the snapshot lacks multi-year publication dates, so recent production cadence cannot be computed. This axis would measure publication frequency and momentum over recent years; the only dated items are the two 2026 arXiv papers.

Reproducibility38/100

Reproducibility scores 38, computed as the mean readiness over 2 papers. Both fall short of a fully reproducible standard in Evolve's review: the Matrix AdaGrad paper is rated medium — theory solid but experiments preliminary and baseline-light — while the world-models survey is rated low, validated only qualitatively. The score could rise quickly if the optimization paper ships complete baselines and the survey adds quantitative benchmarks.

Research summary

Peilin Zhao is a researcher at Shanghai Jiao Tong University whose catalogued work spans two distinct areas: adaptive optimization theory and robotics/embodied AI, with Evolve's bibliometric snapshot splitting his field distribution exactly evenly (50/50) between machine learning (optimization theory) and robotics and embodied AI. The Evolve catalog currently holds two 2026 arXiv papers. The first, Matrix AdaGrad: Row-wise and Column-wise Adaptive Subgradient Methods, places him as last author and derives row- and column-wise adaptive subgradient methods through online mirror descent, with tighter regret guarantees for sparse gradients; Evolve judges the theory solid but the experiments preliminary and baseline-light. The second, World Models for Embodied Intelligence, positions him mid-list on a survey proposing a Plausible→Controllable→Actionable capability ladder for embodied world models — well-argued and formally grounded, though validated only qualitatively. His collaboration footprint includes 17 distinct coauthors (collaboration score 58), with frequent collaborators spanning both fields, from Nanjie Yao and Hao Wang to Li Shen, Gao Huang, Steven Hoi, and Dacheng Tao. Reproducibility is the weakest measured axis (mean 38 across the two papers), consistent with Evolve's readiness ratings of medium and low. Output, impact, and activity scores are not yet computed, so this profile should be read as early-stage: it may change substantially as more of the record is ingested.

Collaboration pattern

Peilin Zhao's catalogued record involves 17 distinct coauthors (collaboration score 58). His frequent-collaborator list — including Nanjie Yao, Hao Wang, Chong Cheng, Zhikang Chen, Jiafei Lyu, Li Shen, Zongqing Lu, Gao Huang, Steven Hoi, and Dacheng Tao — spans both optimization theory and embodied AI. He is not corresponding author on either catalogued paper, sitting last on the AdaGrad paper and mid-list on the survey, a contributing-collaborator pattern rather than a solo-lead one.

Research keywords

adaptive optimizationAdaGradonline mirror descentregret boundsmatrix-valued parametersrow-wise scalingworld modelsembodied intelligencerobot learningmodel-based reinforcement learningvision-language-actioncapability taxonomy

Highlights

Row- and column-wise AdaGrad derived from online mirror descent
As last author, contributed to a principled online-mirror-descent derivation of row- and column-wise adaptive subgradient methods for matrix-valued parameters, with tighter regret bounds for sparse gradients; Evolve rates the theory solid while flagging experiments as preliminary and baseline-light (medium reproducibility readiness).
A Plausible→Controllable→Actionable ladder for embodied world models
Mid-list co-author on a survey that organizes embodied world models into a three-stage capability ladder, from generating plausible scenarios to controllable and finally actionable world models; Evolve finds it well-argued and formally grounded but validated only qualitatively (low reproducibility readiness).
An even 50/50 record across two research fields
The bibliometric field distribution splits exactly 50/50 between machine learning (optimization theory) and robotics/embodied AI, with 17 distinct coauthors bridging both communities — a deliberately broad, early-stage dual-track profile rather than a single-field record.

Timeline

2026Released the arXiv preprint 'Matrix AdaGrad: Row-wise and Column-wise Adaptive Subgradient Methods', listed as last author (not corresponding).
2026Released the arXiv survey 'World Models for Embodied Intelligence: From Plausible to Controllable to Actionable', listed mid-author-list (not corresponding).
2026Catalog snapshot: two arXiv papers, 17 distinct coauthors, and a 50/50 field split between optimization theory and embodied AI.

Papers in the Evolve catalog

Frequent co-authors

Nanjie Yao×1 Hao Wang×1 Chong Cheng×1 Zhikang Chen×1 Wenzhe Li×1 Jiafei Lyu×1 Li Shen×1 Zongqing Lu×1 Gao Huang×1 Steven Hoi×1 Dacheng Tao×1 Deheng Ye×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 scores come from a limited OpenAlex-based snapshot; output, impact, and activity are marked missing rather than estimated.
The profile rests on only two 2026 arXiv papers reviewed by Evolve and may not represent the full publication record.
No ORCID or country is provided and the name is not uncommon; resolution relies on public bibliographic data and identity should be verified before use.
The reproducibility score (38) averages only 2 papers, so it is sensitive to any single new review.