Browsing by Author "Cheng, Wei"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Mastering Long-Tail Complexity on Graphs: Characterization, Learning, and Generalization
    Wang, Haohui; Jing, Baoyu; Ding, Kaize; Zhu, Yada; Cheng, Wei; Zhang, Si; Fan, Yonghui; Zhang, Liqing; Zhou, Dawei (ACM, 2024-08-25)
    In the context of long-tail classification on graphs, the vast majority of existing work primarily revolves around the development of model debiasing strategies, intending to mitigate class imbalances and enhance the overall performance. Despite the notable success, there is very limited literature that provides a theoretical tool for characterizing the behaviors of long-tail classes in graphs and gaining insight into generalization performance in real-world scenarios. To bridge this gap, we propose a generalization bound for long-tail classification on graphs by formulating the problem in the fashion of multi-task learning, i.e., each task corresponds to the prediction of one particular class. Our theoretical results show that the generalization performance of long-tail classification is dominated by the overall loss range and the task complexity. Building upon the theoretical findings, we propose a novel generic framework Hier- Tail for long-tail classification on graphs. In particular, we start with a hierarchical task grouping module that allows us to assign related tasks into hypertasks and thus control the complexity of the task space; then, we further design a balanced contrastive learning module to adaptively balance the gradients of both head and tail classes to control the loss range across all tasks in a unified fashion. Extensive experiments demonstrate the effectiveness of HierTail in characterizing long-tail classes on real graphs, which achieves up to 12.9% improvement over the leading baseline method in balanced accuracy.
  • Thumbnail Image
    The MEso-SCAle Particle Transport model (MESCAPT) for studying sediment dynamics during storms and tsunamis
    Cheng, Wei (Virginia Tech, 2015-12-12)
    Tsunamis and storms are the most devastating coastal hazards that can cause great loss of life and infrastructure damage. To assess tsunami and storm hazard, the magnitude and frequency of each type of event are needed. However, major tsunamis and storms are very infrequent, especially tsunamis, and the only reliable record is the deposits they leave behind. Tsunami and storm deposits can be used to calculate the magnitudes of the respective event, and to contribute to the hazard frequency where there is no historical records. Therefore, for locations where both events could occur, it is crucial to differentiate between the two types of events. Existing studies on the similarities and differences between the two types of deposits all suffer from paucity of the number of events and field data, and a wide range of initial conditions, and thus an unequivocal set of distinguishing deposit characteristics has not been identified yet. In this study, we aim to tackle the problem with the MEso-SCAle Particle Transport model (MESCAPT) that combines the advantages of concentration-based Eulerian methods and particle-based method. The advantage of the former is efficiency and the latter is detailed sediment transport and deposit information. Instead of modeling individual particles, we assume that a group of sediment grains travel and deposit together, which is called a meso-scale particle. This allows simulation domains that are large enough for tsunami and storm wave propagation and inundation. The sediment transport model is coupled with a hydrodynamic model based on the shallow water equations. Simulation results of a case study show good agreements with field measurements of deposits left behind by the 2004 Indian Ocean Tsunami. Idealized tsunami and storm case studies demonstrate the model's capabilities of reproducing morphological changes, as well as microscopic grain-size trends.