1 August 2021 in Journal of Physics: Conference Series
Since the Internet was proposed many years ago, the application of the Internet has become diverse. Now, with the importance of education, the Internet is gradually being used in teaching. Therefore, the purpose of this paper is to explore the application of online and offline teaching mode in embedded system courses under the background of the Internet. In this paper, after consulting the literature on embedded systems and online and offline education models, the data model is constructed by using BP neural network algorithm to process and analyze the relevant data, and the experimental results are drawn. Experimental results show that the Internet-based online and offline teaching model is of some help to the embedded system curriculum.
1 January 2026 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Domain shift, characterized by degraded model performance during the transfer from labeled source domains to unlabeled target domains, poses a persistent challenge for deploying deep learning systems. Current unsupervised domain adaptation (UDA) methods predominantly rely on fine-tuning feature extractors-an approach limited by high computational cost, reduced interpretability, and poor scalability to modern architectures. Our analysis reveals that models pre-trained on large-scale data exhibit domain-invariant geometric patterns in their feature space, characterized by intra-class clustering and inter-class separation, thereby preserving transferable discriminative structures. These findings suggest that cross-domain performance degradation is often associated with decision-boundary misalignment, and that correcting such misalignment can serve as an effective alternative to feature adaptation, particularly when pretrained representations are sufficiently strong. Unlike fine-tuning entire pre-trained models, which risks introducing unpredictable feature distortions, we propose the Feature-space Planes Searcher (FPS): a novel domain adaptation framework that optimizes decision boundaries by leveraging these geometric patterns while keeping the feature encoder frozen. This streamlined approach enables interpretable analysis of adaptation while substantially reducing memory and computational costs through offline feature extraction, permitting full-dataset optimization in a single training cycle. Moreover, we introduce an Intra-Class Distance Metric (ICDM) that enables fully unsupervised hyperparameter selection without requiring target-domain labels. Evaluations on public benchmarks show that FPS achieves competitive performance across standard benchmarks, with notable gains in several settings and tasks. FPS scales efficiently with large multimodal models and shows versatility across diverse domains including protein structure prediction, remote sensing classification, and earthquake detection. We anticipate FPS will provide a simple, effective, and generalizable framework for domain adaptation tasks.
1 February 2021 in IOP Conference Series: Earth and Environmental Science
Employing a temporary array under a high-speed railway viaduct deployed by Peking University, we observe how the spectral characteristics vary with changes of the speed and the model of the train as well as the rail and groundsill by using the clustering algorithm K-Means. For a train in uniform motion, the spectrum of high-speed rail seismic wave is mainly composed of nearly equally spaced peaks and its fundamental frequency is equal to the ratio of the train speed to the carriage length. We reduce the influence from the train speed by aligning the fundamental frequency to make the spectrum pattern clear and easy for comparison. Clustering results show that the spectra of the high-speed rail seismic events have stable patterns under the same train model, rail and groundsill conditions; the stable spectrum patterns change significantly with the changes of the train model, rail and groundsill conditions; monitoring the stable spectral characteristics might be used in safety control of high-speed rail. We apply the clustering method on all the stations of the array. In order to obtain spectra with higher signal to noise ratio at farther stations, we consider the variation of train type and stack the spectrums of three components of high-speed rail seismic signal produced by trains of the same type on the same station. Using the clustering algorithm, we get the regular pattern of how the three component spectra vary with the train type and station position. Based on the above research on the characteristics of high-speed rail seismic spectra and their variation, we propose the concept of 4D ground-frequency map, and discuss its practicability in monitoring the status of high-speed rail and its surrounding media.