In the current landscape of online data services, data transmission and cloud computing are often controlled separately by Internet Service Providers (ISPs) and cloud providers, resulting in significant cooperation challenges and suboptimal global data service optimization. In this study, we propose an end-to-end scheduling method aimed at supporting low-latency and computation-intensive medical services within local wireless networks and healthcare clouds. This approach serves as a practical paradigm for achieving low-latency data services in local private cloud environments. To meet the low-latency requirement while minimizing communication and computation resource usage, we leverage Deep Reinforcement Learning (DRL) algorithms to learn a policy for automatically regulating the transmission rate of medical services and the computation speed of cloud servers. Additionally, we utilize a two-stage tandem queue to address this problem effectively. Extensive experiments are conducted to validate the effectiveness for our proposed method under various arrival rates of medical services.
{"title":"Scheduling of Low-Latency Medical Services in Healthcare Cloud with Deep Reinforcement Learning","authors":"Hongfei Du;Ming Liu;Nianbo Liu;Deying Li;Wenzhong Li;Lifeng Xu","doi":"10.26599/TST.2024.9010033","DOIUrl":"https://doi.org/10.26599/TST.2024.9010033","url":null,"abstract":"In the current landscape of online data services, data transmission and cloud computing are often controlled separately by Internet Service Providers (ISPs) and cloud providers, resulting in significant cooperation challenges and suboptimal global data service optimization. In this study, we propose an end-to-end scheduling method aimed at supporting low-latency and computation-intensive medical services within local wireless networks and healthcare clouds. This approach serves as a practical paradigm for achieving low-latency data services in local private cloud environments. To meet the low-latency requirement while minimizing communication and computation resource usage, we leverage Deep Reinforcement Learning (DRL) algorithms to learn a policy for automatically regulating the transmission rate of medical services and the computation speed of cloud servers. Additionally, we utilize a two-stage tandem queue to address this problem effectively. Extensive experiments are conducted to validate the effectiveness for our proposed method under various arrival rates of medical services.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676358","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169605","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2023.9010148
Jiawei Liu;Haihan Gao;Cheng Yang;Chuan Shi;Tianchi Yang;Hongtao Cheng;Qianlong Xie;Xingxing Wang;Dong Wang
As one of the most crucial topics in the recommendation system field, point-of-interest (POI) recommendation aims to recommending potential interesting POIs to users. Recently, graph neural networks (GNNs) have been successfully used to model interaction and spatio-temporal information in POI recommendations, but the data sparsity of POI recommendations affects the training of GNNs. Although some existing GNN-based POI recommendation approaches try to use social relationships or user attributes to alleviate the data sparsity problem, such auxiliary information is not always available for privacy reasons. Self-supervised learning gives a new idea to alleviate the data sparsity problem, but most existing self-supervised recommendation methods cannot be directly used in the spatio-temporal graph of POI recommendations. In this paper, we propose a novel heterogeneous spatio-temporal graph contrastive learning method, HestGCL, to compensate for existing GNN-based methods' shortcomings. To model spatio-temporal information, we generate spatio-temporally specific views and design view-specific heterogeneous graph neural networks to model spatial and temporal information, respectively. To alleviate data sparsity, we propose a cross-view contrastive strategy to capture differences and correlations among views, providing more supervision signals and boosting the overall performance collaboratively. Extensive experiments on three benchmark datasets demonstrate the effectiveness of HestGCL, which significantly outperforms existing methods.
作为推荐系统领域最重要的课题之一，兴趣点（POI）推荐旨在向用户推荐潜在的感兴趣的兴趣点。最近，图神经网络（GNN）被成功地用于对兴趣点推荐中的交互和时空信息进行建模，但兴趣点推荐的数据稀疏性影响了 GNN 的训练。虽然现有的一些基于 GNN 的 POI 推荐方法试图利用社会关系或用户属性来缓解数据稀疏问题，但出于隐私原因，这些辅助信息并不总是可用的。自监督学习为缓解数据稀疏问题提供了一种新思路，但现有的大多数自监督推荐方法无法直接用于 POI 推荐的时空图。本文提出了一种新型异构时空图对比学习方法 HestGCL，以弥补现有基于 GNN 方法的不足。为了对时空信息建模，我们生成了特定时空的视图，并设计了特定视图的异构图神经网络，分别对空间信息和时间信息建模。为了缓解数据稀缺问题，我们提出了跨视图对比策略，以捕捉视图之间的差异和相关性，从而提供更多的监督信号，协同提升整体性能。在三个基准数据集上的广泛实验证明了 HestGCL 的有效性，其性能明显优于现有方法。
{"title":"Heterogeneous Spatio-Temporal Graph Contrastive Learning for Point-of-Interest Recommendation","authors":"Jiawei Liu;Haihan Gao;Cheng Yang;Chuan Shi;Tianchi Yang;Hongtao Cheng;Qianlong Xie;Xingxing Wang;Dong Wang","doi":"10.26599/TST.2023.9010148","DOIUrl":"https://doi.org/10.26599/TST.2023.9010148","url":null,"abstract":"As one of the most crucial topics in the recommendation system field, point-of-interest (POI) recommendation aims to recommending potential interesting POIs to users. Recently, graph neural networks (GNNs) have been successfully used to model interaction and spatio-temporal information in POI recommendations, but the data sparsity of POI recommendations affects the training of GNNs. Although some existing GNN-based POI recommendation approaches try to use social relationships or user attributes to alleviate the data sparsity problem, such auxiliary information is not always available for privacy reasons. Self-supervised learning gives a new idea to alleviate the data sparsity problem, but most existing self-supervised recommendation methods cannot be directly used in the spatio-temporal graph of POI recommendations. In this paper, we propose a novel heterogeneous spatio-temporal graph contrastive learning method, HestGCL, to compensate for existing GNN-based methods' shortcomings. To model spatio-temporal information, we generate spatio-temporally specific views and design view-specific heterogeneous graph neural networks to model spatial and temporal information, respectively. To alleviate data sparsity, we propose a cross-view contrastive strategy to capture differences and correlations among views, providing more supervision signals and boosting the overall performance collaboratively. Extensive experiments on three benchmark datasets demonstrate the effectiveness of HestGCL, which significantly outperforms existing methods.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676346","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2023.9010095
Zefei Ning;Hao Miao;Zhuolun Jiang;Li Wang
Time series anomaly detection is an important task in many applications, and deep learning based time series anomaly detection has made great progress. However, due to complex device interactions, time series exhibit diverse abnormal signal shapes, subtle anomalies, and imbalanced abnormal instances, which make anomaly detection in time series still a challenge. Fusion and analysis of multivariate time series can help uncover their intrinsic spatio-temporal characteristics, and contribute to the discovery of complex and subtle anomalies. In this paper, we propose a novel approach named Multi-scale Convolution Fusion and Memory-augmented Adversarial AutoEncoder (MCFMAAE) for multivariate time series anomaly detection. It is an encoder-decoder-based framework with four main components. Multi-scale convolution fusion module fuses multi-sensor signals and captures various scales of temporal information. Self-attention-based encoder adopts the multi-head attention mechanism for sequence modeling to capture global context information. Memory module is introduced to explore the internal structure of normal samples, capturing it into the latent space, and thus remembering the typical pattern. Finally, the decoder is used to reconstruct the signals, and then a process is coming to calculate the anomaly score. Moreover, an additional discriminator is added to the model, which enhances the representation ability of autoencoder and avoids overfitting. Experiments on public datasets demonstrate that MCFMAAE improves the performance compared to other state-of-the-art methods, which provides an effective solution for multivariate time series anomaly detection.
{"title":"Using Multi-Scale Convolution Fusion and Memory-Augmented Adversarial Autoencoder to Detect Diverse Anomalies in Multivariate Time Series","authors":"Zefei Ning;Hao Miao;Zhuolun Jiang;Li Wang","doi":"10.26599/TST.2023.9010095","DOIUrl":"https://doi.org/10.26599/TST.2023.9010095","url":null,"abstract":"Time series anomaly detection is an important task in many applications, and deep learning based time series anomaly detection has made great progress. However, due to complex device interactions, time series exhibit diverse abnormal signal shapes, subtle anomalies, and imbalanced abnormal instances, which make anomaly detection in time series still a challenge. Fusion and analysis of multivariate time series can help uncover their intrinsic spatio-temporal characteristics, and contribute to the discovery of complex and subtle anomalies. In this paper, we propose a novel approach named Multi-scale Convolution Fusion and Memory-augmented Adversarial AutoEncoder (MCFMAAE) for multivariate time series anomaly detection. It is an encoder-decoder-based framework with four main components. Multi-scale convolution fusion module fuses multi-sensor signals and captures various scales of temporal information. Self-attention-based encoder adopts the multi-head attention mechanism for sequence modeling to capture global context information. Memory module is introduced to explore the internal structure of normal samples, capturing it into the latent space, and thus remembering the typical pattern. Finally, the decoder is used to reconstruct the signals, and then a process is coming to calculate the anomaly score. Moreover, an additional discriminator is added to the model, which enhances the representation ability of autoencoder and avoids overfitting. Experiments on public datasets demonstrate that MCFMAAE improves the performance compared to other state-of-the-art methods, which provides an effective solution for multivariate time series anomaly detection.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676353","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2023.9010156
Ertong Shang;Hui Liu;Jingyang Zhang;Runqi Zhao;Junzhao Du
Federated learning is an emerging privacy-preserving distributed learning paradigm, in which many clients collaboratively train a shared global model under the orchestration of a remote server. Most current works on federated learning have focused on fully supervised learning settings, assuming that all the data are annotated with ground-truth labels. However, this work considers a more realistic and challenging setting, Federated Semi-Supervised Learning (FSSL), where clients have a large amount of unlabeled data and only the server hosts a small number of labeled samples. How to reasonably utilize the server-side labeled data and the client-side unlabeled data is the core challenge in this setting. In this paper, we propose a new FSSL algorithm for image classification based on consistency regularization and ensemble knowledge distillation, called EKDFSSL. Our algorithm uses the global model as the teacher in consistency regularization methods to enhance both the accuracy and stability of client-side unsupervised learning on unlabeled data. Besides, we introduce an additional ensemble knowledge distillation loss to mitigate model overfitting during server-side retraining on labeled data. Extensive experiments on several image classification datasets show that our EKDFSSL outperforms current baseline methods.
{"title":"Ensemble Knowledge Distillation for Federated Semi-Supervised Image Classification","authors":"Ertong Shang;Hui Liu;Jingyang Zhang;Runqi Zhao;Junzhao Du","doi":"10.26599/TST.2023.9010156","DOIUrl":"https://doi.org/10.26599/TST.2023.9010156","url":null,"abstract":"Federated learning is an emerging privacy-preserving distributed learning paradigm, in which many clients collaboratively train a shared global model under the orchestration of a remote server. Most current works on federated learning have focused on fully supervised learning settings, assuming that all the data are annotated with ground-truth labels. However, this work considers a more realistic and challenging setting, Federated Semi-Supervised Learning (FSSL), where clients have a large amount of unlabeled data and only the server hosts a small number of labeled samples. How to reasonably utilize the server-side labeled data and the client-side unlabeled data is the core challenge in this setting. In this paper, we propose a new FSSL algorithm for image classification based on consistency regularization and ensemble knowledge distillation, called EKDFSSL. Our algorithm uses the global model as the teacher in consistency regularization methods to enhance both the accuracy and stability of client-side unsupervised learning on unlabeled data. Besides, we introduce an additional ensemble knowledge distillation loss to mitigate model overfitting during server-side retraining on labeled data. Extensive experiments on several image classification datasets show that our EKDFSSL outperforms current baseline methods.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676340","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169653","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2023.9010159
Zhiguang Shan;Xu Chen;Yanqiang Zhang;Yifan He;Dandan Wang
Blockchain is one of the most influential technologies in the new round of digital economy development. In order to promote the prosperity of the digital economy with blockchain technology, we need to understand the essence of blockchain and the actual demands of relevant business. This paper delves into the nature of blockchain as a broadcast transmission technology from the perspective of technology evolution and analyzes the necessity of building a blockchain-based public Information Technology (IT) system. In addition, this paper analyzes the architecture, characteristics, and applications regarding trusted public IT system construction by drawing on the design ideas and architecture of Blockchain-based Service Network (BSN).
{"title":"Exploration and Practice of Constructing Trusted Public IT Systems Using Blockchain-Based Service Network","authors":"Zhiguang Shan;Xu Chen;Yanqiang Zhang;Yifan He;Dandan Wang","doi":"10.26599/TST.2023.9010159","DOIUrl":"https://doi.org/10.26599/TST.2023.9010159","url":null,"abstract":"Blockchain is one of the most influential technologies in the new round of digital economy development. In order to promote the prosperity of the digital economy with blockchain technology, we need to understand the essence of blockchain and the actual demands of relevant business. This paper delves into the nature of blockchain as a broadcast transmission technology from the perspective of technology evolution and analyzes the necessity of building a blockchain-based public Information Technology (IT) system. In addition, this paper analyzes the architecture, characteristics, and applications regarding trusted public IT system construction by drawing on the design ideas and architecture of Blockchain-based Service Network (BSN).","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676342","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2024.9010067
Xin Liu;Yi He;Wenxin Tai;Xovee Xu;Fan Zhou;Guangchun Luo
The ability to forecast future events brings great benefits for society and cyberspace in many public safety domains, such as civil unrest, pandemics and crimes. The occurrences of new events are often correlated or dependent on historical and concurrent events. Many existing studies learn event-occurring processes with sequential and structural models, which, however, suffer from inefficient and inaccurate prediction problems. To better understand the event forecasting task and characterize the occurrence of new events, we exploit the human cognitive theory from the cognitive neuroscience discipline to find available cues for algorithm design and event prediction. Motivated by the dual process theory, we propose a two-stage learning scheme for event knowledge mining and prediction. First, we screen out event candidates based on historical inherent knowledge. Then we re-rank event candidates by probing into the newest relative events. Our proposed model mimics a sociological phenomenon called “the chameleon effect” and consists of a new target attentive graph collaborative learning mechanism to ensure a better understanding of sophisticated evolution patterns associated with events. In addition, self-supervised contrastive learning is employed to alleviate the over-smoothing problem that existed in graph learning while improving the model's interpretability. Experiments show the effectiveness of our approach.
{"title":"Exploring the Chameleon Effect of Contextual Dynamics in Temporal Knowledge Graph for Event Prediction","authors":"Xin Liu;Yi He;Wenxin Tai;Xovee Xu;Fan Zhou;Guangchun Luo","doi":"10.26599/TST.2024.9010067","DOIUrl":"https://doi.org/10.26599/TST.2024.9010067","url":null,"abstract":"The ability to forecast future events brings great benefits for society and cyberspace in many public safety domains, such as civil unrest, pandemics and crimes. The occurrences of new events are often correlated or dependent on historical and concurrent events. Many existing studies learn event-occurring processes with sequential and structural models, which, however, suffer from inefficient and inaccurate prediction problems. To better understand the event forecasting task and characterize the occurrence of new events, we exploit the human cognitive theory from the cognitive neuroscience discipline to find available cues for algorithm design and event prediction. Motivated by the dual process theory, we propose a two-stage learning scheme for event knowledge mining and prediction. First, we screen out event candidates based on historical inherent knowledge. Then we re-rank event candidates by probing into the newest relative events. Our proposed model mimics a sociological phenomenon called “the chameleon effect” and consists of a new target attentive graph collaborative learning mechanism to ensure a better understanding of sophisticated evolution patterns associated with events. In addition, self-supervised contrastive learning is employed to alleviate the over-smoothing problem that existed in graph learning while improving the model's interpretability. Experiments show the effectiveness of our approach.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676361","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As the device complexity keeps increasing, the blockchain networks have been celebrated as the cornerstone of numerous prominent platforms owing to their ability to provide distributed and immutable ledgers and data-driven autonomous organizations. The distributed consensus algorithm is the core component that directly dictates the performance and properties of blockchain networks. However, the inherent characteristics of the shared wireless medium, such as fading, interference, and openness, pose significant challenges to achieving consensus within these networks, especially in the presence of malicious jamming attacks. To cope with the severe consensus problem, in this paper, we present a distributed jamming-resilient consensus algorithm for blockchain networks in wireless environments, where the adversary can jam the communication channel by injecting jamming signals. Based on a non-binary slight jamming model, we propose a distributed four-stage algorithm to achieve consensus in the wireless blockchain network, including leader election, leader broadcast, leader aggregation, and leader announcement stages. With high probability, we prove that our jamming-resilient algorithm can ensure the validity, agreement, termination, and total order properties of consensus with the time complexity of $O(n)$ . Both theoretical analyses and empirical simulations are conducted to verify the consistency and efficiency of our algorithm.
{"title":"Jamming-Resilient Consensus for Wireless Blockchain Networks","authors":"Yifei Zou;Meng Hou;Li Yang;Minghui Xu;Libing Wu;Dongxiao Yu;Xiuzhen Cheng","doi":"10.26599/TST.2023.9010160","DOIUrl":"https://doi.org/10.26599/TST.2023.9010160","url":null,"abstract":"As the device complexity keeps increasing, the blockchain networks have been celebrated as the cornerstone of numerous prominent platforms owing to their ability to provide distributed and immutable ledgers and data-driven autonomous organizations. The distributed consensus algorithm is the core component that directly dictates the performance and properties of blockchain networks. However, the inherent characteristics of the shared wireless medium, such as fading, interference, and openness, pose significant challenges to achieving consensus within these networks, especially in the presence of malicious jamming attacks. To cope with the severe consensus problem, in this paper, we present a distributed jamming-resilient consensus algorithm for blockchain networks in wireless environments, where the adversary can jam the communication channel by injecting jamming signals. Based on a non-binary slight jamming model, we propose a distributed four-stage algorithm to achieve consensus in the wireless blockchain network, including leader election, leader broadcast, leader aggregation, and leader announcement stages. With high probability, we prove that our jamming-resilient algorithm can ensure the validity, agreement, termination, and total order properties of consensus with the time complexity of \u0000<tex>$O(n)$</tex>\u0000. Both theoretical analyses and empirical simulations are conducted to verify the consistency and efficiency of our algorithm.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676355","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169667","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Action segmentation has made significant progress, but segmenting and recognizing actions from untrimmed long videos remains a challenging problem. Most state-of-the-art methods focus on designing models based on temporal convolution. However, the limitations of modeling long-term temporal dependencies and the inflexibility of temporal convolutions restrict the potential of these models. To address the issue of over-segmentation in existing action segmentation methods, which leads to classification errors and reduced segmentation quality, this paper proposes a global spatial-temporal information encoder-decoder based action segmentation method. The method proposed in this paper uses the global temporal information captured by refinement layer to assist the Encoder-Decoder (ED) structure in judging the action segmentation point more accurately and, at the same time, suppress the excessive segmentation phenomenon caused by the ED structure. The method proposed in this paper achieves 93% frame accuracy on the constructed real Tai Chi action dataset. The experimental results prove that this method can accurately and efficiently complete the long video action segmentation task.
动作分割技术已经取得了重大进展，但从未经剪辑的长视频中分割和识别动作仍然是一个具有挑战性的问题。大多数最先进的方法都侧重于设计基于时态卷积的模型。然而，长期时间依赖性建模的局限性和时间卷积的不灵活性限制了这些模型的潜力。为了解决现有动作分割方法中存在的过度分割问题，即导致分类错误和分割质量下降的问题，本文提出了一种基于全局时空信息编码器-解码器的动作分割方法。本文提出的方法利用细化层捕获的全局时空信息，辅助编码器-解码器（ED）结构更准确地判断动作分割点，同时抑制 ED 结构造成的过度分割现象。本文提出的方法在构建的真实太极拳动作数据集上实现了 93% 的帧准确率。实验结果证明，该方法可以准确高效地完成长视频动作分割任务。
{"title":"Global Spatial-Temporal Information Encoder-Decoder Based Action Segmentation in Untrimmed Video","authors":"Yichao Liu;Yiyang Sun;Zhide Chen;Chen Feng;Kexin Zhu","doi":"10.26599/TST.2024.9010041","DOIUrl":"https://doi.org/10.26599/TST.2024.9010041","url":null,"abstract":"Action segmentation has made significant progress, but segmenting and recognizing actions from untrimmed long videos remains a challenging problem. Most state-of-the-art methods focus on designing models based on temporal convolution. However, the limitations of modeling long-term temporal dependencies and the inflexibility of temporal convolutions restrict the potential of these models. To address the issue of over-segmentation in existing action segmentation methods, which leads to classification errors and reduced segmentation quality, this paper proposes a global spatial-temporal information encoder-decoder based action segmentation method. The method proposed in this paper uses the global temporal information captured by refinement layer to assist the Encoder-Decoder (ED) structure in judging the action segmentation point more accurately and, at the same time, suppress the excessive segmentation phenomenon caused by the ED structure. The method proposed in this paper achieves 93% frame accuracy on the constructed real Tai Chi action dataset. The experimental results prove that this method can accurately and efficiently complete the long video action segmentation task.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676351","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169649","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-11DOI: 10.26599/TST.2023.9010149
Xin Xi;Jinhui Yuan;Shan Lu;Jieyue He
The combination of multiple drugs is a significant therapeutic strategy that can enhance treatment effectiveness and reduce medication side effects. However, identifying effective synergistic drug combinations in a vast search space remains challenging. Current methods for predicting synergistic drug combinations primarily rely on calculating drug similarity based on the drug heterogeneous network or drug information, enabling the prediction of pairwise synergistic drug combinations. However, these methods not only fail to fully study the rich information in drug heterogeneous networks, but also can only predict pairwise drug combinations. To address these limitations, we present a novel Synergistic Multi-drug Combination prediction method of Western medicine based on Heterogeneous Network representation learning with Contrastive Learning, called SMC-HNCL. Specifically, two drug features are learnt from different perspectives using the drug heterogeneous network and anatomical therapeutic chemical (ATC) codes, and fused by attention mechanism. Furthermore, a group representation method based on multi-head self-attention is employed to learn representations of drug combinations, innovatively realizing the prediction of synergistic multi-drug combinations. Experimental results demonstrate that SMC-HNCL outperforms the state-of-the-art baseline methods in predicting synergistic drug pairs on two synergistic drug combination datasets and can also effectively predict synergistic multi-drug combinations.
{"title":"Synergistic Multi-Drug Combination Prediction Based on Heterogeneous Network Representation Learning with Contrastive Learning","authors":"Xin Xi;Jinhui Yuan;Shan Lu;Jieyue He","doi":"10.26599/TST.2023.9010149","DOIUrl":"https://doi.org/10.26599/TST.2023.9010149","url":null,"abstract":"The combination of multiple drugs is a significant therapeutic strategy that can enhance treatment effectiveness and reduce medication side effects. However, identifying effective synergistic drug combinations in a vast search space remains challenging. Current methods for predicting synergistic drug combinations primarily rely on calculating drug similarity based on the drug heterogeneous network or drug information, enabling the prediction of pairwise synergistic drug combinations. However, these methods not only fail to fully study the rich information in drug heterogeneous networks, but also can only predict pairwise drug combinations. To address these limitations, we present a novel Synergistic Multi-drug Combination prediction method of Western medicine based on Heterogeneous Network representation learning with Contrastive Learning, called SMC-HNCL. Specifically, two drug features are learnt from different perspectives using the drug heterogeneous network and anatomical therapeutic chemical (ATC) codes, and fused by attention mechanism. Furthermore, a group representation method based on multi-head self-attention is employed to learn representations of drug combinations, innovatively realizing the prediction of synergistic multi-drug combinations. Experimental results demonstrate that SMC-HNCL outperforms the state-of-the-art baseline methods in predicting synergistic drug pairs on two synergistic drug combination datasets and can also effectively predict synergistic multi-drug combinations.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676350","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169604","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The data asset is emerging as a crucial component in both industrial and commercial applications. Mining valuable knowledge from the data benefits decision-making and business. However, the usage of data assets raises tension between sensitive information protection and value estimation. As an emerging machine learning paradigm, Federated Learning (FL) allows multiple clients to jointly train a global model based on their data without revealing it. This approach harnesses the power of multiple data assets while ensuring their privacy. Despite the benefits, it relies on a central server to manage the training process and lacks quantification of the quality of data assets, which raises privacy and fairness concerns. In this work, we present a novel framework that combines Federated Learning and Blockchain by Shapley value (FLBS) to achieve a good trade-off between privacy and fairness. Specifically, we introduce blockchain in each training round to elect aggregation and evaluation nodes for training, enabling decentralization and contribution-aware incentive distribution, with these nodes functionally separated and able to supervise each other. The experimental results validate the effectiveness of FLBS in estimating contribution even in the presence of heterogeneity and noisy data.
数据资产正在成为工业和商业应用中的重要组成部分。从数据中挖掘有价值的知识有利于决策和业务。然而，数据资产的使用引发了敏感信息保护与价值评估之间的矛盾。作为一种新兴的机器学习范式，联合学习（FL）允许多个客户在不泄露数据的情况下，基于其数据共同训练一个全局模型。这种方法既能利用多种数据资产的力量，又能确保其隐私。尽管好处多多，但它依赖于一个中央服务器来管理训练过程，缺乏对数据资产质量的量化，从而引发了对隐私和公平性的担忧。在这项工作中，我们提出了一个新颖的框架，将联邦学习（Federated Learning）和夏普利区块链（Blockchain by Shapley value，FLBS）结合起来，在隐私和公平性之间实现了良好的权衡。具体来说，我们在每一轮训练中引入区块链，选举出训练的聚合节点和评估节点，实现去中心化和贡献感知的激励分配，这些节点在功能上相互分离并能够相互监督。实验结果验证了 FLBS 在估计贡献方面的有效性，即使在存在异质性和噪声数据的情况下也是如此。
{"title":"Quantifying Bytes: Understanding Practical Value of Data Assets in Federated Learning","authors":"Minghao Yao;Saiyu Qi;Zhen Tian;Qian Li;Yong Han;Haihong Li;Yong Qi","doi":"10.26599/TST.2024.9010034","DOIUrl":"https://doi.org/10.26599/TST.2024.9010034","url":null,"abstract":"The data asset is emerging as a crucial component in both industrial and commercial applications. Mining valuable knowledge from the data benefits decision-making and business. However, the usage of data assets raises tension between sensitive information protection and value estimation. As an emerging machine learning paradigm, Federated Learning (FL) allows multiple clients to jointly train a global model based on their data without revealing it. This approach harnesses the power of multiple data assets while ensuring their privacy. Despite the benefits, it relies on a central server to manage the training process and lacks quantification of the quality of data assets, which raises privacy and fairness concerns. In this work, we present a novel framework that combines Federated Learning and Blockchain by Shapley value (FLBS) to achieve a good trade-off between privacy and fairness. Specifically, we introduce blockchain in each training round to elect aggregation and evaluation nodes for training, enabling decentralization and contribution-aware incentive distribution, with these nodes functionally separated and able to supervise each other. The experimental results validate the effectiveness of FLBS in estimating contribution even in the presence of heterogeneity and noisy data.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":null,"pages":null},"PeriodicalIF":6.6,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10676341","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142169641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：