Existing fusion strategies often involve simple concatenation of modal features,disregarding persona-lized fusion requirements based on the characteristics of each modality.Additionally,solely considering the emo-tions of individual utterances in isolation,without accounting for their emotional states within the context,can lead to errors in emotion recognition.To address the aforementioned issues,this paper proposes a multimodal emotion recognition method based on a layered fusion strategy and the incorporation of contextual information.The method employs a layered fusion strategy,progressively integrating different modal features in a hierarchical manner to re-duce noise interference from individual modalities and address inconsistencies in expression across different mo-dalities.It leverages the contextual information to comprehensively analyze the emotional representation of each utterance within the context,enhancing overall emotion recognition performance.In binary emotion classification tasks,the proposed method achieves a 1.54%improvement in accuracy compared with the state-of-the-art(SOTA)model.In multi-class emotion recognition tasks,the F1 score is improved by 2.79%compared to SOTA model.
In botnet attacks,because the characteristics of disguised botnet traffic data are too similar to normal traffic data,it is difficult to distinguish them accurately by traditional detection methods.In order to solve this problem,this paper proposes a Multi-feature Aggregation Spectral Graph Wavelet Neural Network(MFA-SGWNN).This method combines the attribute and spatial features of traffic,which can effectively capture the hidden characteristics of infected host traffic,enhance the feature representation of botnet nodes,and avoid the influence of unbalanced data samples and malicious encrypted traffic on detection.Experimental results on the ISCX2014 botnet and CIC-IDS 2017(botnet)datasets show that MFA-SGWNN outperforms existing methods and has stronger robustness and generalization ability.
Among various approaches of few-shot named entity recognition(NER),two-stage models based on prototype networks are widely used.However,these methods can not fully utilize the semantic information in entity labels and overly relies on entity type prototype vectors in distance calculation,resulting in poor generalization ability of the model.To address these issues,this paper proposes a few-shot named entity recognition method based on label semantic information awareness.This method consists of a two-stage process:entity span detection and entity type classification.When constructing entity type prototype vectors,the semantic information associated with the corresponding entity types is considered and fused with the prototype vectors through a dimension transformation layer.During the entity recognition of new samples,entity type positive and negative samples are combined with entity type prototype vectors to form entity type triplets,and the samples are classified based on the distance to the triplets.Experimental results on multiple datasets demonstrate that the proposed model significantly outperforms previous models.
The mid-Holocene (MH) is the most recent typical climate period and a subject of great interest in global paleocultural research. Following the latest Paleoclimate Modeling Intercomparison Project phase 4 (PMIP4) protocol and using a fully coupled climate model, we simulated the climate during both the MH and the preindustrial (PI) periods and quantified the effects of Earth orbital parameters (ORBs) and greenhouse gases (GHGs) on climate differences, focusing on the simulated differences in the Atlantic meridional overturning circulation (AMOC) between these two periods. Compared to the PI simulation, the ORB effect in the MH simulation led to seasonal enhancement of temperature, consistent with previous findings. In the MH simulation, the ORB effect led to a markedly warmer climate in the mid–high latitudes and increased precipitation in the Northern Hemisphere, which were partially offset by the cooling effect of the lower GHGs. The AMOC in the MH simulation was about 4 % stronger than that in the PI simulation. The ORB effect led to 6 % enhancement of the AMOC in the MH simulation, which was, however, partly neutralized by the GHG effect. Transient simulation from the MH to the PI further demonstrated the opposite effects of ORBs and GHGs on the evolution of the AMOC during the past 6000 years. The simulated stronger AMOC in the MH was mainly due to the thinner sea ice in the polar oceans caused by the ORB effect, which reduced the freshwater flux export to the subpolar Atlantic and resulted in a more saline North Atlantic. This study may help us quantitatively understand the roles of different external forcing factors in Earth's climate evolution since the MH.
A generative adversarial network（GAN） model is applied to the outpainting reconstruction on the micro grain and pore structure of sandstone thin-section image, and the semantics of the predicted image is analyzed in the model. The findings demonstrate that the model can predict a lager view of sandstone microstructure, which is 2.25 times the size of the original vision, and has good performance for different types of rock image semantics. The predicted image semantics, such as surface texture of different grains, grain morphology and complex contact relationship between multiple grains are consistent with the real results. However, in the task of outpainting reconstruction on microscopic special phenomenon in rock thin-section, the model lacks sensitivity to special phenomenon. In the task of reconstruction on pore structure, the prediction error of micropore distribution is larger than that of common pore spaces such as intergranular pores, fractures and dissolution pores.The prediction performance of reconstruction results of different pore spaces may be related to pore features（such as pore size and connectivity）.
