Siyi Yu

Siyi Yu

Auto-generated profile, based on publicly available information

Siyi Yu

Shanghai University of Sport, Shanghai, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 April 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
SYSiyi Yu
Siyi Yu
XZXizhi Zhao
Xizhi Zhao
JMJinlong Ma
Jinlong Ma
YLYujian Liu
Yujian Liu
The Winter Olympic Games stimulate significant infrastructure development, yet post-Games venue utilization remains a pressing challenge. Traditional survey-based assessments are often resource-intensive and lack the capacity for continuous, large-scale monitoring. This study employs nighttime light (NTL) remote sensing data to analyze 209 Winter Olympic Game venues, utilizing time series analysis, change point detection, and seasonal decomposition. Results indicate that 32.06% of venues exhibit a significant increase in activity, while 12.92% experience a decline significantly. Two distinct utilization patterns emerge: outdoor sports venues for pre-2000 Games are more likely to face post-Games underutilization due to geographic isolation and high maintenance costs, whereas urban indoor venues can sustain long-term use through multi-purpose adaptability and accessibility. The findings confirm NTL data as an effective real-time monitoring tool, supporting data-driven legacy planning. This study highlights the significance of multi-purpose design, tourism integration, and policy-driven repurpose in ensuring the long-term sustainability of Olympic infrastructure investments.

28 November 2023
Verified
Conference Paper
Closed access
SYSiyi Yu
Siyi Yu
WLWei Luo
Wei Luo
Gang LiGang Li
Gang Li
BYBo Yang
Bo Yang
Few-shot learning (FSL) aims to learn to new concepts based on very limited data. One of the main challenges in FSL is the use of pretrained embeddings whose dimension is too high for the small sample size. While the large dimensionality helps capture diverse features in the training example, not all the features are relevant to the task at hand. Such task-irrelevant features from pre-trained embeddings can negatively impact the performance of the FSL model. To address this challenge, we propose a simple yet effective way to learn task-level manifolds that explicitly eliminate the negative effect of nuisance features. One key novelty of our approach is to address class imbalance in the training data by injecting supervised information through key virtual samples, which we named Manifold Support Points. Empirically, we demonstrate that the stability of the learnt task-level manifolds and their effectiveness in challenging tasks in FSL. Our approach shows competitive performance on several benchmark datasets, including tieredImageNet and CUB.