Attention-Enhanced Voice Portrait Model Using Generative Adversarial Network

Voice portrait technology has explored and established the relationship between speakers’ voices and their facialfeatures, aiming to generate corresponding facial characteristics by providing the voice of an unknown speaker.Due to its powerful advantages in image generation, Generative Adversarial Networks (GANs) have now beenwidely applied across various fields. The existing Voice2Face methods for voice portraits are primarily based onGANs trained on voice-face paired datasets. However, voice portrait models solely constructed on GANs facelimitations in image generation quality and struggle to maintain facial similarity. Additionally, the training processis relatively unstable, thereby affecting the overall generative performance of the model. To overcome the abovechallenges,wepropose a novel deepGenerativeAdversarialNetworkmodel for audio-visual synthesis, namedAVPGAN(Attention-enhanced Voice Portrait Model using Generative Adversarial Network). This model is based ona convolutional attention mechanism and is capable of generating corresponding facial images from the voice ofan unknown speaker. Firstly, to address the issue of training instability, we integrate convolutional neural networkswith deep GANs. In the network architecture, we apply spectral normalization to constrain the variation of thediscriminator, preventing issues such as mode collapse. Secondly, to enhance the model’s ability to extract relevantfeatures between the two modalities, we propose a voice portrait model based on convolutional attention. Thismodel learns the mapping relationship between voice and facial features in a common space from both channeland spatial dimensions independently. Thirdly, to enhance the quality of generated faces, we have incorporated adegradation removal module and utilized pretrained facial GANs as facial priors to repair and enhance the clarityof the generated facial images. Experimental results demonstrate that our AVP-GAN achieved a cosine similarity of0.511, outperforming the performance of our comparison model, and effectively achieved the generation of highqualityfacial images corresponding to a speaker’s voice.

An End-To-End Hyperbolic Deep Graph Convolutional Neural Network Framework

Graph Convolutional Neural Networks(GCNs)have been widely used in various fields due to their powerful capabilities in processing graph-structured data.However,GCNs encounter significant challenges when applied to scale-free graphs with power-law distributions,resulting in substantial distortions.Moreover,most of the existing GCN models are shallow structures,which restricts their ability to capture dependencies among distant nodes and more refined high-order node features in scale-free graphs with hierarchical structures.To more broadly and precisely apply GCNs to real-world graphs exhibiting scale-free or hierarchical structures and utilize multi-level aggregation of GCNs for capturing high-level information in local representations,we propose the Hyperbolic Deep Graph Convolutional Neural Network(HDGCNN),an end-to-end deep graph representation learning framework that can map scale-free graphs from Euclidean space to hyperbolic space.In HDGCNN,we define the fundamental operations of deep graph convolutional neural networks in hyperbolic space.Additionally,we introduce a hyperbolic feature transformation method based on identity mapping and a dense connection scheme based on a novel non-local message passing framework.In addition,we present a neighborhood aggregation method that combines initial structural featureswith hyperbolic attention coefficients.Through the above methods,HDGCNN effectively leverages both the structural features and node features of graph data,enabling enhanced exploration of non-local structural features and more refined node features in scale-free or hierarchical graphs.Experimental results demonstrate that HDGCNN achieves remarkable performance improvements over state-ofthe-art GCNs in node classification and link prediction tasks,even when utilizing low-dimensional embedding representations.Furthermore,when compared to shallow hyperbolic graph convolutional neural network models,HDGCNN exhibits notable advantages and performance enhancements.

Angle-resolved photoemission study of NbGeSb with non-symmorphic symmetry

We investigate the electronic structure of NbGeSb with non-symmorphic symmetry.We employ angle-resolved photoemission spectroscopy(ARPES)to observe and identify the bulk and surface states over the Brillouin zone.By utilizing high-energy photons,we identify the bulk Fermi surface and bulk nodal line along the direction X–R,while the Fermi surface of the surface state is observed by using low-energy photons.We observe the splitting of surface bands away from the high-symmetry point X.The density functional theory calculations on bulk and 1 to 5-layer slab models,as well as spin textures of NbGeSb,verify that the band splitting could be attributed to the Rashba-like spin–orbit coupling caused by space-inversion-symmetry breaking at the surface.These splitted surface bands cross with each other,forming two-dimensional Weyl-like crossings that are protected by mirror symmetry.Our findings provide insights into the two-dimensional topological and symmetry-protected band inversion of surface states.

Followers of the Social Media Network and National Security

With the social media networks development quickly, the followers of the social media network＇ s behaviors have taken a lot of damagers and threats to national security, and made the nation into unstable situations, even subverted the national sovereignty .This paper analyzes the characters of the followers of the social media in the Ira, Tunisia, Egypt and Libya＇s turmoil, concludes constructive suggestions how to ensure national stability and harmonious development, has some positive effect to our national security.

Chinese Cyber Threat Intelligence Named Entity Recognition via RoBERTa-wwm-RDCNN-CRF

In recent years,cyber attacks have been intensifying and causing great harm to individuals,companies,and countries.The mining of cyber threat intelligence(CTI)can facilitate intelligence integration and serve well in combating cyber attacks.Named Entity Recognition(NER),as a crucial component of text mining,can structure complex CTI text and aid cybersecurity professionals in effectively countering threats.However,current CTI NER research has mainly focused on studying English CTI.In the limited studies conducted on Chinese text,existing models have shown poor performance.To fully utilize the power of Chinese pre-trained language models(PLMs)and conquer the problem of lengthy infrequent English words mixing in the Chinese CTIs,we propose a residual dilated convolutional neural network(RDCNN)with a conditional random field(CRF)based on a robustly optimized bidirectional encoder representation from transformers pre-training approach with whole word masking(RoBERTa-wwm),abbreviated as RoBERTa-wwm-RDCNN-CRF.We are the first to experiment on the relevant open source dataset and achieve an F1-score of 82.35%,which exceeds the common baseline model bidirectional encoder representation from transformers(BERT)-bidirectional long short-term memory(BiLSTM)-CRF in this field by about 19.52%and exceeds the current state-of-the-art model,BERT-RDCNN-CRF,by about 3.53%.In addition,we conducted an ablation study on the encoder part of the model to verify the effectiveness of the proposed model and an in-depth investigation of the PLMs and encoder part of the model to verify the effectiveness of the proposed model.The RoBERTa-wwm-RDCNN-CRF model,the shared pre-processing,and augmentation methods can serve the subsequent fundamental tasks such as cybersecurity information extraction and knowledge graph construction,contributing to important applications in downstream tasks such as intrusion detection and advanced persistent threat(APT)attack detection.

Multi-Branch Deepfake Detection Algorithm Based on Fine-Grained Features

With the rapid development of deepfake technology,the authenticity of various types of fake synthetic content is increasing rapidly,which brings potential security threats to people’s daily life and social stability.Currently,most algorithms define deepfake detection as a binary classification problem,i.e.,global features are first extracted using a backbone network and then fed into a binary classifier to discriminate true or false.However,the differences between real and fake samples are often subtle and local,and such global feature-based detection algorithms are not optimal in efficiency and accuracy.To this end,to enhance the extraction of forgery details in deep forgery samples,we propose a multi-branch deepfake detection algorithm based on fine-grained features from the perspective of fine-grained classification.First,to address the critical problem in locating discriminative feature regions in fine-grained classification tasks,we investigate a method for locating multiple different discriminative regions and design a lightweight feature localization module to obtain crucial feature representations by augmenting the most significant parts of the feature map.Second,using information complementation,we introduce a correlation-guided fusion module to enhance the discriminative feature information of different branches.Finally,we use the global attention module in the multi-branch model to improve the cross-dimensional interaction of spatial domain and channel domain information and increase the weights of crucial feature regions and feature channels.We conduct sufficient ablation experiments and comparative experiments.The experimental results show that the algorithm outperforms the detection accuracy and effectiveness on the FaceForensics++and Celeb-DF-v2 datasets compared with the representative detection algorithms in recent years,which can achieve better detection results.

CDR2IMG:A Bridge from Text to Image in Telecommunication Fraud Detection

Telecommunication fraud has run rampant recently worldwide.However,previous studies depend highly on expert knowledge-based feature engineering to extract behavior information,which cannot adapt to the fastchanging modes of fraudulent subscribers.Therefore,we propose a new taxonomy that needs no hand-designed features but directly takes raw Call DetailRecords(CDR)data as input for the classifier.Concretely,we proposed a fraud detectionmethod using a convolutional neural network(CNN)by taking CDR data as images and applying computer vision techniques like image augmentation.Comprehensive experiments on the real-world dataset from the 2020 Digital Sichuan Innovation Competition show that our proposed method outperforms the classic methods in many metrics with excellent stability in both the changes of quantity and the balance of samples.Compared with the state-of-the-art method,the proposed method has achieved about 89.98%F1-score and 92.93%AUC,improving 2.97%and 0.48%,respectively.With the augmentation technique,the model’s performance can be further enhanced by a 91.09%F1-score and a 94.49%AUC respectively.Beyond telecommunication fraud detection,our method can also be extended to other text datasets to automatically discover new features in the view of computer vision and its powerful methods.

Visualizing the electronic structure of kagome magnet LuMn6Sn6 by angle-resolved photoemission spectroscopy

Searching for the dispersionless flat band(FB)in quantum materials,especially in topological systems,becomes an interesting topic.The kagome lattice is an ideal platform for such exploration because the FB can be naturally induced by the underlying destructive interference.Nevertheless,the magnetic kagome system that hosts the FB close to the Fermi level(EF)is exceptionally rare.Here,we study the electronic structure of a kagome magnet LuMn6Sn6 by combining angleresolved photoemission spectroscopy and density functional theory calculations.The observed Fermi-surface topology and overall band dispersions are similar to previous studies of the XMn6Sn6(X=Dy,Tb,Gd,Y)family of compounds.We clearly observe two kagome-derived FBs extending through the entire Brillouin zone,and one of them is located just below EF.The photon-energy-dependent measurements reveal that these FBs are nearly dispersionless along the kz direction as well,supporting the quasi-two-dimensional character of such FBs.Our results complement the XMn6Sn6 family and demonstrate the robustness of the FB features across this family.

Quantitative analysis of school safety events in China

Recent years have seen increasing school safety events along with growing numbers of students in China.In this paper,more than 400 serious school safety events between 2000 and 2018 in China were collected.The causes and characteristics of these events,taking into account the occurrence years,months,regions,education stages and types,were investigated.The results indicate that the number of school safety events has generally increased annually from 2000 to 2018 and can significantly vary each month,showing a higher frequency of occurrence during the“First Semester”(generally from September to December in China).Moreover,spatial distribution of school safety events is normally related to regional economic development;it was found that Guangdong,Jiangsu and Shandong is a first-level occurrence hotspots,followed by Zhejiang,Henan,Hebei and Sichuan,which are secondary occurrence hotspots.Furthermore,statistical analysis shows that the number of school safety events that occurred in kindergartens,primary schools and middle schools are approximately equal(around 1/3).Finally,it was found that the school safety events caused by“Accident”(such as school bus accidents,school fires,crowded stampedes and the collapses of school buildings)occupy a large proportion(57%).