Yizhou Yu

University of Hong Kong · Hong Kong SAR China

Unclaimed · AI-structured profileOpenAlex

University of Hong Kong researcher: sparse, parameter-efficient adaptation for continual learning and video diffusion

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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 cannot be scored: the OpenAlex record provides no reliable publication-volume series (h-index and i10-index are both 0), so total output, venue quality, and authorship-role distribution cannot yet be measured for this researcher.

Impact—

Impact cannot be scored: the record shows h-index 0 and i10-index 0 with no citation counts by year, so citation volume, influence trajectory, and scholarly uptake remain unmeasured rather than genuinely zero.

Focus0/100

The focus score is 0, reflecting extreme topical dispersion rather than missing data: the OpenAlex topic profile spreads across five clusters — Advanced Neural Network Applications (6), Domain Adaptation and Few-Shot Learning (4), GANs and Image Synthesis (3), Multimodal Machine Learning (3), Reinforcement Learning in Robotics (2) — and field distribution splits exactly 50/50 between computer vision (video generation) and machine learning (computer vision). Even so, the three reviewed 2026 papers cluster tightly on efficient adaptation and generative-model diagnosis.

Collaboration48/100

Collaboration scores 48/100, a mid-range reading: the record lists 10 distinct coauthors, with a stable circle including Chaoqi Chen, Yunxiang Fu, and Meng Lou. Authorship placement points to a senior-position but non-corresponding role — last author on both continual-learning papers (CLARE; One Adapter, Many Tasks) and middle author on the video-diffusion study, with no corresponding-author flag in the catalog.

Activity64/100

Activity scores 64/100. The record counts 8 works in the recent three-year window, and the per-year series places all 8 in 2026 — a sharply concentrated burst that includes three arXiv preprints reviewed by Evolve. Output before 2026 is absent from this snapshot, so the activity reading reflects a current surge rather than a sustained multi-year cadence; no ORCID linkage is available to corroborate it.

Reproducibility55/100

Reproducibility scores 55/100, computed from 2 scored papers; all three reviewed preprints carry a medium repro-readiness rating. Reviewer verdicts flag concrete gaps: CLARE's strong 100-task class-incremental gains lack a mixture-of-experts baseline; the video-diffusion study reports marginal gains with no external baselines; One Adapter, Many Tasks shows reporting inconsistencies despite broad evidence — all limiting independent replication confidence.

Research summary

Yizhou Yu is a researcher at the University of Hong Kong whose recent record concentrates on continual learning for vision models and on diagnosing video diffusion transformers. Three 2026 arXiv preprints reviewed by Evolve form the core of the catalog: CLARE, a sparsity-based class-incremental learning framework combining L1-learned sparse masks with capacity-aware freezing; One Adapter, Many Tasks, which shows a single task-conditioned adapter outperforming per-task and LoRA-merge baselines at much lower cost; and Temporal State Transport, a spectral-analysis study of temporal-attention imbalance in video diffusion with a training-free correction. The OpenAlex topic profile is consistent with this spread — Advanced Neural Network Applications (6), Domain Adaptation and Few-Shot Learning (4), GANs and Image Synthesis (3), Multimodal ML (3), RL in Robotics (2) — and field distribution splits evenly (50/50) between computer vision (video generation) and machine learning (computer vision). Activity is front-loaded: 8 works in the last three years, all recorded in 2026, with 10 distinct coauthors. On the reviewed papers Yu appears as last author twice and middle author once, never as corresponding author. Reproducibility is mid-range (mean 55 across 2 scored papers), with reviewers noting missing baselines and reporting inconsistencies. Citation-level impact cannot yet be assessed from the available record.

Collaboration pattern

Yu works in compact co-author teams around a stable 10-person circle (frequent partners include Chaoqi Chen, Yunxiang Fu, Meng Lou, and Luyao Tang), with 10 distinct coauthors in the OpenAlex record and a mid-range collaboration score of 48/100. On all three Evolve-reviewed 2026 papers Yu takes a non-corresponding role — last author on the two continual-learning papers and middle author on the video-diffusion study — consistent with a senior contributor within an ongoing group rather than a solo corresponding lead.

Research keywords

class-incremental learningcontinual learningcatastrophic forgettingsparsityparameter-efficient fine-tuningadapterslow-rank adaptation (LoRA)video diffusion modelstemporal attentionattention entropyspectral analysistraining-free enhancement

Highlights

CLARE: sparsity-based scaling for 100-task class-incremental learning
On the reviewed paper, L1-learned sparse masks plus capacity-aware freezing of a shared adapter deliver strong gains as tasks scale to 100; Evolve notes the claim is tempered by a missing mixture-of-experts baseline.
One task-conditioned adapter replaces per-task adapters at lower cost
The reviewed paper shows a single task-conditioned adapter beating per-task and LoRA-merge class-incremental baselines at far lower cost, with broad evidence across settings though some reporting inconsistencies remain.
Spectral diagnosis of temporal imbalance in video diffusion
The reviewed paper proposes spectral-tension-guided query temperature as a training-free correction for spectral imbalance in video diffusion; Evolve calls the diagnostic elegant and the analysis thorough, but gains are marginal with no external baselines.

Timeline

2026Publishes One Adapter, Many Tasks, showing a single task-conditioned adapter beating per-task and LoRA-merge continual-learning baselines at far lower cost (Evolve-reviewed arXiv preprint; last author).
2026Publishes Temporal State Transport, a spectral analysis of temporal-attention imbalance in video diffusion with a training-free correction (middle author).
2026Publishes CLARE, a sparsity-based class-incremental learning framework with strong results in the 100-task setting (last author).
2026OpenAlex snapshot logs 8 works in the recent three-year window, all in 2026, with 10 distinct coauthors; h- and i10-index read 0, indicating incomplete citation coverage.

Papers in the Evolve catalog

Frequent co-authors

Yunxiang Fu×2 Meng Lou×2 Luyao Tang×1 Bingjun Luo×1 Dong Yi×1 Jialin Guo×1 Haoning Xi×1 Cheng Chen×1 Chaoqi Chen×1 Zicheng Liao×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 an OpenAlex-resolved record only; h-index and i10-index read 0, likely reflecting incomplete attribution rather than genuinely zero impact.
Yizhou Yu is a relatively common name; this profile is name-resolved from public bibliographic data without ORCID linkage, so records may be merged or missing.
All three cataloged papers are 2026 arXiv preprints; peer-reviewed venue status is not captured in the available data.
Output and impact dimensions are unscored because the record lacks citation counts and a per-year publication series.