Knowledge distillation is an effective method for training lightweight vision models. However, acquiring teacher supervision for training samples is often costly, especially from large-scale models like vision transformers (ViTs). In this paper, we develop a simple framework to reduce the supervision cost of ViT distillation: masking out a fraction of input tokens given to the teacher. By masking input tokens, one can skip the computations associated with the masked tokens without requiring any change to teacher parameters or architecture. We find that masking patches with the lowest student attention scores is highly effective, saving up to 50% of teacher FLOPs without any drop in student accuracy, while other masking criterion leads to suboptimal efficiency gains. Through in-depth analyses, we reveal that the student-guided masking provides a good curriculum to the student, making teacher supervision easier to follow during the early stage and challenging in the later stage.
Sangjoon Park, Yongsung Kwon, Hyungjoon Soh et al.
Predicting temporal patterns across various domains poses significant challenges due to their nuanced and often nonlinear trajectories. To address this challenge, prediction frameworks have been continuously refined, employing data-driven statistical methods, mathematical models, and machine learning. Recently, as one of the challenging systems, shared transport systems such as public bicycles have gained prominence due to urban constraints and environmental concerns. Predicting rental and return patterns at bicycle stations remains a formidable task due to the system's openness and imbalanced usage patterns across stations. In this study, we propose a deep learning framework to predict rental and return patterns by leveraging cartogram approaches. The cartogram approach facilitates the prediction of demand for newly installed stations with no training data as well as long-period prediction, which has not been achieved before. We apply this method to public bicycle rental-and-return data in Seoul, South Korea, employing a spatial-temporal convolutional graph attention network. Our improved architecture incorporates batch attention and modified node feature updates for better prediction accuracy across different time scales. We demonstrate the effectiveness of our framework in predicting temporal patterns and its potential applications.
Seong-Gyu Yang, Beom Jun Kim, Seung-Woo Son et al.
Complex network analyses have provided clues to improve power-grid stability with the help of numerical models. The high computational cost of numerical simulations, however, has inhibited the approach, especially when it deals with the dynamic properties of power grids such as frequency synchronization. In this study, we investigate machine learning techniques to estimate the stability of power-grid synchronization. We test three different machine learning algorithms -- random forest, support vector machine, and artificial neural network -- training them with two different types of synthetic power grids consisting of homogeneous and heterogeneous input-power distribution, respectively. We find that the three machine learning models better predict the synchronization stability of power-grid nodes when they are trained with the heterogeneous input-power distribution than the homogeneous one. With the real-world power grids of Great Britain, Spain, France, and Germany, we also demonstrate that the machine learning algorithms trained on synthetic power grids are transferable to the stability prediction of the real-world power grids, which implies the prospective applicability of machine learning techniques on power-grid studies.