Early exiting is an effective paradigm for improving the inference efficiency of pre-trained language models (PLMs) by dynamically adjusting the number of executed layers for each sample. However, in most existing works, easy and hard samples are treated equally by each classifier during training, which neglects the test-time early exiting behavior, leading to inconsistency between training and testing. Although some methods have tackled this issue under a fixed speed-up ratio, the challenge of flexibly adjusting the speed-up ratio while maintaining consistency between training and testing is still under-explored. To bridge the gap, we propose a novel Consistency-Oriented Signal-based Early Exiting (COSEE) framework, which leverages a calibrated sample weighting mechanism to enable each classifier to emphasize the samples that are more likely to exit at that classifier under various acceleration scenarios. Extensive experiments on the GLUE benchmark demonstrate the effectiveness of our COSEE across multiple exiting signals and backbones, yielding a better trade-off between performance and efficiency.

Early exiting has demonstrated great potential in accelerating the inference of pre-trained language models (PLMs) by enabling easy samples to exit at shallow layers, eliminating the need for executing deeper layers. However, existing early exiting methods primarily rely on class-relevant logits to formulate their exiting signals for estimating prediction certainty, neglecting the detrimental influence of class-irrelevant information in the features on prediction certainty. This leads to an overestimation of prediction certainty, causing premature exiting of samples with incorrect early predictions. To remedy this, we define an NSP score to estimate prediction certainty by considering the proportion of class-irrelevant information in the features. On this basis, we propose a novel early exiting method based on the Certainty-Aware Probability (CAP) score, which integrates insights from both logits and the NSP score to enhance prediction certainty estimation, thus enabling more reliable exiting decisions. The experimental results on the GLUE benchmark show that our method can achieve an average speed-up ratio of 2.19× across all tasks with negligible performance degradation, surpassing the state-of-the-art (SOTA) ConsistentEE by 28%, yielding a better trade-off between task performance and inference efficiency. The code is available at https://github.com/He-Jianing/NSP.git.

The information collection in the global governance of the digital economy is important to companies in response to COVID-19. This research studied the initiative of conflicts between privacy and information disclosure based on agency theory, analyzed the resolution of the conflicts based on incentive compatibility, and further discussed the rationale of the balance between private and public interest based on agile governance. This research suggests the necessity of finding the balance between public interest and privacy protection based on the hierarchical division of private and public interest. The dynamic psychological behavior to privacy and information disclosure by uninfected and infected citizens in response to COVID-19 is simulated by Volterra differential equations. The specification of boundaries in data use can be helpful to companies in reconciling the privacy and information disclosure for customer relationship management in digital governance in response to COVID-19.

Multivariate time series prediction task has long been a crucial problem in machine learning since it includes applications in real life such as energy consumption, traffic congestion and economic trending. Challenge is that there are many factors affecting the accuracy of prediction, such as information in time domain and frequency domain, as well as interdependence among sequences. Therefore, effectively capturing and combining features in different domains becomes the focus of research. In this paper, a deep learning framework enhanced by wavelet transform is proposed to predict multivariate time series. Our model captures the long-term and short-term characteristics of time series in the time domain respectively through convolutional network. Further more, we use LSTM to achieve trending vector of short-term features, while for long-term features, we use the multi-head attention mechanism to get its representation. For frequency-domain features, our framework extracts from the scalo-gram generated by applying wavelet transform to the time sequence. Through the design of CNN convolution kernel, our model also ensures that the dependence between time series can be extracted. Experimental results show that our proposed model outperforms other baseline approaches on three of the four realworld datasets with sophisticated patterns.

In recent years, Ethernet passive optical network (EPON) has been considered as a promising candidate of mobile front-haul networks, since it can provide high capacity for bandwidth-hungry enhanced mobile broadband (eMBB) traffic. However, the EPON cannot support ultra-reliable and low latency communication (uRLLC) traffic, which is sporadic, highly unpredictable, and delay-sensitive. To address this problem, this paper devises a hybrid dynamic bandwidth allocation (DBA) algorithm, which treats the uRLLC traffic as high-priority traffic. In particular, the hybrid DBA algorithm schedules the uRLLC traffic in a competitive manner every other fixed time interval and polls the eMBB traffic in a round-robin manner. Our simulation results show that the hybrid DBA algorithm can work very well in the network with 16 ONUs.

As a famous cultural wealth, the sketch of fresco is one of the most important art expression forms in the World Heritage. To avoid the damage of natural and human factors, painters can only use photos and videos to depict sketch in most world culture heritage sites, otherwise real frescos. Therefore, a computational method for extracting sketch is helpful and meaningful in sketch copying and researching area. However, existing approaches with the incomplete fresco are not enough to deal with the challenge of sketch extraction. In this paper, we proposes a framework to generate sketch of fresco with frescos as input. To reduce noise and refine detail lines, we adopt hierarchical segmentation technology to extract sketches of different regions respectively, and splice those sketches into an integrated sketch. For replacing the missing content and getting a complete sketch, we provide recommendations from a database of many existing fresco sketches elements for users to select. At last, users can adjust sketch based on vectorization to achieve optimization. Our framework can be used for artists or learners to study painting sketch images and research fresco art. The experimental results demonstrate the effectiveness of our framework.

The study of digital Chinese calligraphy has become more valuable nowadays. While existing researches on Chinese calligraphy analysis are primarily focused on stroke-based character recognition and simulation, we propose a global feature descriptor to deal with style recognition problem in this paper. The proposed method extracts three categories of character features: position features, proportion features and projection features. These features are then used to train an SVM classifier of calligraphy style. We test the global feature based classifier on five-style Chinese calligraphy character set. The experimental results show that the proposed method can achieve good classification accuracy, proving the effectiveness of the global feature descriptor in calligraphy style recognition.