Pengfei Zhou

National University of Singapore

Unclaimed · AI-structured profileOpenAlex

NUS researcher across efficient multimodal learning, game AI, and adaptive vision token pruning

papers here

reviewed by Evolve

first-author

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—

Output data is insufficient in this snapshot: no publication counts or venue record are available yet. This axis would measure total publication volume, venue quality, and authorship roles across the researcher's full career record.

Impact—

Impact data is unavailable in this snapshot: no citation counts or h-index are provided. This axis would measure citations, uptake, and the influence of the reviewed work across the community over time.

Focus0/100

The focus axis scores 0, computed directly from the field distribution: Zhou's cataloged output splits exactly one-third each across game AI and interactive systems, multimodal AI systems, and computer vision / efficient multimodal inference. With no field holding a plurality, this reads as deliberately breadth-first — two surveys set taxonomies in different areas while one methods paper targets efficient inference — rather than an absence of direction. The score reflects balance across three fields, not scattered output.

Collaboration69/100

Collaboration scores 69, backed by a raw count of 31 distinct coauthors in the snapshot. All three cataloged papers are middle-position, non-corresponding contributions, consistent with working inside larger teams rather than leading small groups. A recurring core — including Meng Luo, Yanlin Li, Hao Li, Hongzhan Lin, Tianjie Ju, Ran Zhang, Yeying Jin, Mong-Li Lee, Wynne Hsu, Pan Wang, Siwei Song, and Hui Ji — suggests stable project clusters around the NUS affiliation.

Activity—

Activity data is insufficient in this snapshot: a recent-year publication cadence cannot be computed from the available records. This axis would measure ongoing output rhythm and research momentum across the most recent publication years.

Reproducibility43/100

Reproducibility scores 43, a mean computed across the three Evolve-reviewed papers. Readiness ratings split between medium for the efficient multimodal learning survey and the vision token pruning routing paper, and low for the game AI survey, which introduces no experiments and rests on a fast-moving preprint base. The routing paper carries the strongest empirical footing, with +26.9% relative accuracy on VTC-Bench Group A, though real-world routing remains untested.

Research summary

Pengfei Zhou is a researcher at the National University of Singapore whose Evolve-reviewed work spans three interleaved areas: efficient multimodal learning, game AI in the foundation-model era, and adaptive inference for multimodal large language models. The bibliometric field distribution splits evenly — one third each across game AI and interactive systems, multimodal AI systems, and computer vision / efficient multimodal inference — which yields a focus score of 0 and marks this as an early-stage, breadth-first profile rather than one anchored to a single subfield. All three catalog papers are 2026 arXiv preprints with Zhou in a middle authorship position and none as corresponding author. Two are surveys: a comprehensive review of efficient multimodal learning claiming the first Model-Algorithm-System taxonomy (covering model compression, KV cache management, and mixture-of-experts), and a six-role survey of AI across the game lifecycle. The third is a methods paper proposing sample-adaptive strategy routing for vision token pruning, reporting +26.9% relative accuracy over the best fixed strategy on VTC-Bench Group A — solid on its tested subset, though real-world routing remains untested. Collaboration data records 31 distinct coauthors and a collaboration score of 69, with a recurring core that includes Mong-Li Lee, Wynne Hsu, Tianjie Ju, and other named collaborators. Reproducibility scores 43 across the three reviewed papers, while output, impact, and activity remain unscored for lack of data.

Collaboration pattern

The snapshot records 31 distinct coauthors and a collaboration score of 69, with all three cataloged papers written from middle authorship positions and none as corresponding author. The recurring core spans at least twelve names — Meng Luo, Yanlin Li, Hao Li, Hongzhan Lin, Tianjie Ju, Ran Zhang, Yeying Jin, Mong-Li Lee, Wynne Hsu, Pan Wang, Siwei Song, and Hui Ji — indicating team-based projects at NUS rather than solo or small-group work.

Research keywords

efficient multimodal learningmultimodal large language models (MLLMs)vision token pruningadaptive inference and strategy routingmodel compressionKV cache managementmixture-of-expertsgame AIworld modelsLLM agentsprocedural content generationsurvey and taxonomy building

Highlights

First claimed Model-Algorithm-System taxonomy for efficient multimodal learning
Evolve's review credits this 2026 survey with the first Model-Algorithm-System taxonomy for efficient multimodal learning, calling it coherent and well-sourced across model compression, KV cache management, and mixture-of-experts, while noting that it offers no quantitative synthesis.
Six-role map of AI across the game lifecycle
This 2026 survey organizes game AI in the foundation-model era into six roles spanning the game lifecycle, with a strong taxonomy and calibrated evidence reading; Evolve notes it adds no new experiments and rests on a fast-moving preprint base.
Sample-adaptive routing for vision token pruning
The paper proposes routing per sample among pruning strategies for multimodal large language models, reporting +26.9% relative accuracy over the best fixed strategy on VTC-Bench Group A; Evolve judges it solid on its tested subset, with real-world routing still untested.

Timeline

2026Co-authored a comprehensive survey of efficient multimodal learning, claiming the first Model-Algorithm-System taxonomy.
2026Co-authored a six-role survey of AI for games across the game lifecycle in the foundation model era.
2026Co-authored the sample-adaptive strategy routing method for vision token pruning in multimodal large language models.

Papers in the Evolve catalog

Frequent co-authors

Meng Luo×1 Yanlin Li×1 Hao Li×1 Hongzhan Lin×1 Tianjie Ju×1 Ran Zhang×1 Yeying Jin×1 Mong-Li Lee×1 Wynne Hsu×1 Pan Wang×1 Siwei Song×1 Hui Ji×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.

Profile is name-resolved from public bibliographic data; no ORCID is provided, and 'Pengfei Zhou' is a relatively common name, so the identity match to a specific person should be verified.
All dimension scores come from a single bibliometric snapshot; the output, impact, and activity axes lack sufficient data and are left unscored.
The three reviewed papers are 2026 arXiv preprints; no peer-reviewed venue information is included in the snapshot.
No citation counts, h-index, or full publication history are provided; coverage is limited to Evolve-reviewed papers and basic bibliometric indicators.