Search for: All records

Machine unlearning is becoming increasingly important as deep models become more prevalent, particularly when there are frequent requests to remove the influence of specific training data due to privacy concerns or erroneous sensing signals. Spatial-temporal Graph Neural Networks, in particular, have been widely adopted in real-world applications that demand efficient unlearning, yet research in this area remains in its early stages. In this paper, we introduce STEPS, a framework specifically designed to address the challenges of spatio-temporal graph unlearning. Our results demonstrate that STEPS not only ensures data continuity and integrity but also significantly reduces the time required for unlearning, while minimizing the accuracy loss in the new model compared to a model with 0% unlearning. 

Abstract Municipalities worldwide implement stormwater management programs to mitigate the hydrological and water quality impacts of stormwater runoff. Stormwater utility fees (SWUF) are often used to fund such important programs by collecting revenue from residential and commercial properties. However, existing SWUFs often solely rely on the estimate of impervious surfaces and do not consider other environmental, infrastructure, and socioeconomic factors in the generation and effects of stormwater runoff. This study is the first attempt to propose a reconstruction of SWUFs from the perspectives of social equity and environmental justice. The method aims to address disparities in fee rates among residential parcels, focusing on helping economically disadvantaged communities. It integrates drainage service, potential contribution to non-point source pollution, and socioeconomic status through two alternative schemes. The two schemes allocate fees based on combined rankings of the three factors at the level of census block groups. The proposed method was applied to 88 180 residential parcels in Corpus Christi, Texas, a mid-sized coastal community. The results suggest that over 70% of the disadvantaged communities would benefit from the reconstructed SWUFs without affecting the targeted funding for stormwater infrastructure. This method builds on publicly available datasets and offers an adaptive framework for other municipalities to incorporate additional factors or datasets, representing an exploratory step toward achieving more equitable stormwater management practices. 

Abstract The quantification of urban impervious area has important implications for the design and management of urban water and environmental infrastructure systems. This study proposes a deep learning model to classify 15‐cm aerial imagery of urban landscapes, coupled with a vector‐oriented post‐classification processing algorithm for automatically retrieving canopy‐covered impervious surfaces. In a case study in Corpus Christi, TX, deep learning classification covered an area of approximately 312 km²(or 14.86 billion 0.15‐m pixels), and the post‐classification effort led to the retrieval of over 4 km²(or 0.18 billion pixels) of additional impervious area. The results also suggest the underestimation of urban impervious area by existing methods that cannot consider the canopy‐covered impervious surfaces. By improving the identification and quantification of various impervious surfaces at the city scale, this study could directly benefit a variety of environmental and infrastructure management practices and enhance the reliability and accuracy of processed‐based models for urban hydrology and water infrastructure.