Electronic registration identification technology (ERI) has developed rapidly in recent years. This technology has been widely used in large urban transportation monitoring, vehicle counting, identification, and traffic congestion detection. It has many advantages, such as long recognition distance, high recognition accuracy, more information stored, fast reading speed, etc. Currently this technology has achieved full coverage of the entire road network and vehicles in the cities where it is applied. Despite the richness of this data, there are significant limitations in terms of vehicle trajectories, especially in terms of spatial and temporal density. Compared with ERI trajectories, vehicle GPS trajectories have a higher sampling rate, but we are unable to obtain more comprehensive and complete vehicle GPS data due to the limitations of vehicle technology and security factors. In this paper, we innovatively propose a new method to reconstruct fine-grained ERI trajectories by learning from taxi GPS data. This approach can be divided into two steps. First, a novel Taxi-ERI traffic network is proposed to connect ERI data and taxi data. It’s a directed multi-graph whose nodes are consisted of all ERI acquisition points and edges are composed of clustered taxi trajectories. Then, the probability of each road is calculated by a Bayes classification based on the multi-road travel time distribution model while there are multi roads between two adjacent acquisition points, the model parameters are trained by the expectation maximization (EM) algorithm. Finally, we extensively evaluate the proposed framework on the taxi trajectory dataset and ERI data collected from Chongqing, China. The experimental results show that the method can accurately reconstruct vehicle trajectories.
{"title":"Fine-grained Reconstruction of Vehicle Trajectories Based on Electronic Registration Identification Data","authors":"Xin Chen, Linjiang Zheng, Wengang Li, Longquan Liao, Qixing Wang, Xingze Yang","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00065","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00065","url":null,"abstract":"Electronic registration identification technology (ERI) has developed rapidly in recent years. This technology has been widely used in large urban transportation monitoring, vehicle counting, identification, and traffic congestion detection. It has many advantages, such as long recognition distance, high recognition accuracy, more information stored, fast reading speed, etc. Currently this technology has achieved full coverage of the entire road network and vehicles in the cities where it is applied. Despite the richness of this data, there are significant limitations in terms of vehicle trajectories, especially in terms of spatial and temporal density. Compared with ERI trajectories, vehicle GPS trajectories have a higher sampling rate, but we are unable to obtain more comprehensive and complete vehicle GPS data due to the limitations of vehicle technology and security factors. In this paper, we innovatively propose a new method to reconstruct fine-grained ERI trajectories by learning from taxi GPS data. This approach can be divided into two steps. First, a novel Taxi-ERI traffic network is proposed to connect ERI data and taxi data. It’s a directed multi-graph whose nodes are consisted of all ERI acquisition points and edges are composed of clustered taxi trajectories. Then, the probability of each road is calculated by a Bayes classification based on the multi-road travel time distribution model while there are multi roads between two adjacent acquisition points, the model parameters are trained by the expectation maximization (EM) algorithm. Finally, we extensively evaluate the proposed framework on the taxi trajectory dataset and ERI data collected from Chongqing, China. The experimental results show that the method can accurately reconstruct vehicle trajectories.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80500193","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The goal of few-shot semantic segmentation (FSS) is to segment the foreground image of an unseen class in the query image by using a few labeled support images. Existing two-branch models mine support and query image information to improve segmentation results by employing support prototypes, calculating the similarity between support and query images, or fusing multi-scale features. Such methods only focus on the spatial information of the query image in the initial feature extraction and subsequent processes. Meanwhile, limited by the sample size, their ability to extract channel information is insufficient, thus leading to the information loss of the query image. To solve the issues, we propose an implicit channel relation based few-shot semantic segmentation method entitled MANGO. The implicit relation mining process is implemented after the initial feature extraction and before the two branches interact to fully mine the query image information. Specifically, the query channel features are taken as nodes to construct the graph structure to establish the relationship between nodes. The network motif is used to quantity the attribute features and structural features of nodes to enhance the relationship between channels. Finally, we aggregate the two features and mine the implicit relationship of nodes through graph representation learning. Experiments on PASCAL-5i and FSS-1000 datasets demonstrate that our proposed method outperforms the state-of-the-art methods.
{"title":"Mining Implicit Relations Among Image Channels for Few-Shot Semantic Segmentation","authors":"Xu Yuan, Ying Yang, Huafei Huang, Shuo Yu, Lili Cong","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00062","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00062","url":null,"abstract":"The goal of few-shot semantic segmentation (FSS) is to segment the foreground image of an unseen class in the query image by using a few labeled support images. Existing two-branch models mine support and query image information to improve segmentation results by employing support prototypes, calculating the similarity between support and query images, or fusing multi-scale features. Such methods only focus on the spatial information of the query image in the initial feature extraction and subsequent processes. Meanwhile, limited by the sample size, their ability to extract channel information is insufficient, thus leading to the information loss of the query image. To solve the issues, we propose an implicit channel relation based few-shot semantic segmentation method entitled MANGO. The implicit relation mining process is implemented after the initial feature extraction and before the two branches interact to fully mine the query image information. Specifically, the query channel features are taken as nodes to construct the graph structure to establish the relationship between nodes. The network motif is used to quantity the attribute features and structural features of nodes to enhance the relationship between channels. Finally, we aggregate the two features and mine the implicit relationship of nodes through graph representation learning. Experiments on PASCAL-5i and FSS-1000 datasets demonstrate that our proposed method outperforms the state-of-the-art methods.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79406077","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00318
Yi Yang, Wenqiang Ma, Wenqiao Sun, Haibin Zhang t, Zhiqiang Liu, Lexi Xu, Ye Zhu
As an emerging technology, digital twin (DT) has great potential to address the challenges posed by the dynamics and complexity of vehicles in vehicular edge computing (VEC) networks. By mapping the VEC network to the virtual space, DT can monitor vehicles, road side units (RSUs), channels, and resource usage in real time, further bringing comprehensive and accurate network analysis to the VEC network. However, the real-world implement of DT-empowered VEC networks cannot avoid the collection of privacy-sensitive information of participants. An incentive mechanism is necessitated to identify the qualities of participants without prior information and incent them to participate in DT modeling, so as to realize the requirement of privacy preserving while improving the DT modeling efficiency. In this paper, We propose a combined multi-armed bandit-based auction (CMABA) incentive mechanism that can identify the quality of clients in the VEC network without revealing sensitive and private information, and achieve the optimal performance of the model under budget constraints. The simulation results show that this scheme can significantly incent high-quality clients to participate in DT modeling under the requirement of privacy preserving and the constraint of limited budget, and improve the accuracy of DT modeling.
{"title":"Privacy-Preserving Digital Twin for Vehicular Edge Computing Networks","authors":"Yi Yang, Wenqiang Ma, Wenqiao Sun, Haibin Zhang t, Zhiqiang Liu, Lexi Xu, Ye Zhu","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00318","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00318","url":null,"abstract":"As an emerging technology, digital twin (DT) has great potential to address the challenges posed by the dynamics and complexity of vehicles in vehicular edge computing (VEC) networks. By mapping the VEC network to the virtual space, DT can monitor vehicles, road side units (RSUs), channels, and resource usage in real time, further bringing comprehensive and accurate network analysis to the VEC network. However, the real-world implement of DT-empowered VEC networks cannot avoid the collection of privacy-sensitive information of participants. An incentive mechanism is necessitated to identify the qualities of participants without prior information and incent them to participate in DT modeling, so as to realize the requirement of privacy preserving while improving the DT modeling efficiency. In this paper, We propose a combined multi-armed bandit-based auction (CMABA) incentive mechanism that can identify the quality of clients in the VEC network without revealing sensitive and private information, and achieve the optimal performance of the model under budget constraints. The simulation results show that this scheme can significantly incent high-quality clients to participate in DT modeling under the requirement of privacy preserving and the constraint of limited budget, and improve the accuracy of DT modeling.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84964705","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00327
Feiyang Tang
Biometric data privacy is becoming a major concern for many organizations in the age of big data, particularly in the ICT sector, because it may be easily exploited in apps. Most apps utilize biometrics by accessing common application programming interfaces (APIs); hence, we aim to categorize their usage. The categorization based on behavior may be closely correlated with the sensitive processing of a user’s biometric data, hence highlighting crucial biometric data privacy assessment concerns. We propose PABAU, Privacy Analysis of Biometric API Usage. PABAU learns semantic features of methods in biometric APIs and uses them to detect and categorize the usage of biometric API implementation in the software according to their privacy-related behaviors. This technique bridges the communication and background knowledge gap between technical and non-technical individuals in organizations by providing an automated method for both parties to acquire a rapid understanding of the essential behaviors of biometric API in apps, as well as future support to data protection officers (DPO) with legal documentation, such as conducting a Data Protection Impact Assessment (DPIA).
{"title":"PABAU: Privacy Analysis of Biometric API Usage","authors":"Feiyang Tang","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00327","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00327","url":null,"abstract":"Biometric data privacy is becoming a major concern for many organizations in the age of big data, particularly in the ICT sector, because it may be easily exploited in apps. Most apps utilize biometrics by accessing common application programming interfaces (APIs); hence, we aim to categorize their usage. The categorization based on behavior may be closely correlated with the sensitive processing of a user’s biometric data, hence highlighting crucial biometric data privacy assessment concerns. We propose PABAU, Privacy Analysis of Biometric API Usage. PABAU learns semantic features of methods in biometric APIs and uses them to detect and categorize the usage of biometric API implementation in the software according to their privacy-related behaviors. This technique bridges the communication and background knowledge gap between technical and non-technical individuals in organizations by providing an automated method for both parties to acquire a rapid understanding of the essential behaviors of biometric API in apps, as well as future support to data protection officers (DPO) with legal documentation, such as conducting a Data Protection Impact Assessment (DPIA).","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85011289","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00340
Yingying Yao, Xiaolin Chang, Lin Li, Jiqiang Liu, J. Misic, V. Mišić
The recent advances of emerging technologies including artificial intelligence, 5G, 6G, extended reality and blockchain promote the rapid development of next-generation Internet. As an evolving paradigm of next-generation Internet, metaverse, a fully immersive, hyper spatiotemporal and selfsustaining virtual shared space, is moving from imagination to the coming reality. However, its massive data flow, pervasive user profiling activities and other intrinsic features can lead to a lot of security and privacy concerns, which will hinder its further deployment. Specially, since the identities of users/avatars in the metaverse can be illegally stolen, impersonated, and interoperability issues can be encountered in authentication across metaverses, this paper designs a lightweight and privacy-preserving seamless cross-metaverse authentication and key agreement scheme named MetaverseAKA to meet the challenges. Metaverse-AKA can not only realize the seamless cross-metaverse authentication but also assure the users’ privacy by achieving the anonymity and unlinkability. In addition, Metaverse-AKA also has the following advantages: (i) Realizing the traceability for users in physical world. (ii) Resistance to multiple attacks like impersonation attack, man-in-the-middle attack and replay attack. (iii) Adopting lightweight cryptographic prinitives and having better performance through experiment verification and comparison.
{"title":"Metaverse-AKA: A Lightweight and PrivacyPreserving Seamless Cross-Metaverse Authentication and Key Agreement Scheme","authors":"Yingying Yao, Xiaolin Chang, Lin Li, Jiqiang Liu, J. Misic, V. Mišić","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00340","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00340","url":null,"abstract":"The recent advances of emerging technologies including artificial intelligence, 5G, 6G, extended reality and blockchain promote the rapid development of next-generation Internet. As an evolving paradigm of next-generation Internet, metaverse, a fully immersive, hyper spatiotemporal and selfsustaining virtual shared space, is moving from imagination to the coming reality. However, its massive data flow, pervasive user profiling activities and other intrinsic features can lead to a lot of security and privacy concerns, which will hinder its further deployment. Specially, since the identities of users/avatars in the metaverse can be illegally stolen, impersonated, and interoperability issues can be encountered in authentication across metaverses, this paper designs a lightweight and privacy-preserving seamless cross-metaverse authentication and key agreement scheme named MetaverseAKA to meet the challenges. Metaverse-AKA can not only realize the seamless cross-metaverse authentication but also assure the users’ privacy by achieving the anonymity and unlinkability. In addition, Metaverse-AKA also has the following advantages: (i) Realizing the traceability for users in physical world. (ii) Resistance to multiple attacks like impersonation attack, man-in-the-middle attack and replay attack. (iii) Adopting lightweight cryptographic prinitives and having better performance through experiment verification and comparison.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85092428","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00048
Fengyuan Zhang, Zhiwen Yu, Yimeng Liu, Helei Cui, Bin Guo
Mobile crowdsensing (MCS) collects sensing data by recruiting task participants to realize large-scale sensing tasks in cities. However, due to the limitations of human activity range and sensing mode, relying only on human participants to achieve this process will lead to sensing blind areas, ultimately affecting the integrity and validity of sensing data. With the rise of unmanned vehicles (UVs) and sensor-assisted MCS research, it provides new inspirations for solving complex sensing tasks in smart cities. In this article, we propose heterogeneous crowdsensing, which includes heterogeneous participants such as human participants, UVs, and fixed sensors. Our goal is to accomplish large-scale, high-quality urban sensing tasks by collaborating with these three types of heterogeneous participants. To solve the collaborative sensing problem, we propose an algorithm called spatio-temporal PPO (STPPO). We first define the capability and cost attributes of the heterogeneous participants and then divide the large-scale sensing area into a set of subregions by a subgraph construction method. Based on the spatio-temporal characteristics of the subregions and the attributes of the heterogeneous participants, we finally solve the cooperative scheduling problem of the subregions using proximal policy optimization (PPO) algorithms to maximize the overall POI collection rate and collection fairness. Finally, extensive experiments are conducted based on real datasets. The overall results of STPPO outperform other baselines, with a 30.19% performance improvement compared to the PPO algorithm.
{"title":"Spatio-temporal Feature Based Multi-participant Recruitment in Heterogeneous Crowdsensing","authors":"Fengyuan Zhang, Zhiwen Yu, Yimeng Liu, Helei Cui, Bin Guo","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00048","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00048","url":null,"abstract":"Mobile crowdsensing (MCS) collects sensing data by recruiting task participants to realize large-scale sensing tasks in cities. However, due to the limitations of human activity range and sensing mode, relying only on human participants to achieve this process will lead to sensing blind areas, ultimately affecting the integrity and validity of sensing data. With the rise of unmanned vehicles (UVs) and sensor-assisted MCS research, it provides new inspirations for solving complex sensing tasks in smart cities. In this article, we propose heterogeneous crowdsensing, which includes heterogeneous participants such as human participants, UVs, and fixed sensors. Our goal is to accomplish large-scale, high-quality urban sensing tasks by collaborating with these three types of heterogeneous participants. To solve the collaborative sensing problem, we propose an algorithm called spatio-temporal PPO (STPPO). We first define the capability and cost attributes of the heterogeneous participants and then divide the large-scale sensing area into a set of subregions by a subgraph construction method. Based on the spatio-temporal characteristics of the subregions and the attributes of the heterogeneous participants, we finally solve the cooperative scheduling problem of the subregions using proximal policy optimization (PPO) algorithms to maximize the overall POI collection rate and collection fairness. Finally, extensive experiments are conducted based on real datasets. The overall results of STPPO outperform other baselines, with a 30.19% performance improvement compared to the PPO algorithm.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85235727","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00196
Xin Wang, Hongbin Shi
Multi-hop knowledge base question answering aims to answer natural language questions through multi-hop relation reasoning in the knowledge base. An important challenge of the task is the lack of labels for reasoning paths, which leads to the possibility to produce correct answers through incorrect paths in the training, and cannot generalize well in testing. Recently research has attempted to handle the challenge by devising reward shaping or introducing additional information to generate supervision signals of intermediate paths. But they required extra expert experience and label information. To address this situation, we propose a novel method under the teacher-student framework, it leverages perturbation consistency to learn intermediate paths. In the teacher network, we construct close data points for intermediate path prediction by applying random perturbations. Inspired by the data smoothing assumption that labels of close data points should be the same, a consistency loss over predictions of constructed data points and original ones is evaluated. The student network is used to answer questions more precisely by leveraging the intermediate distribution learned from the teacher network. Extensive experiments on two benchmark datasets are conducted, and the results have demonstrated the effectiveness of the proposed method.
{"title":"Leveraging Perturbation Consistency to Improve Multi-hop Knowledge Base Question Answering","authors":"Xin Wang, Hongbin Shi","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00196","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00196","url":null,"abstract":"Multi-hop knowledge base question answering aims to answer natural language questions through multi-hop relation reasoning in the knowledge base. An important challenge of the task is the lack of labels for reasoning paths, which leads to the possibility to produce correct answers through incorrect paths in the training, and cannot generalize well in testing. Recently research has attempted to handle the challenge by devising reward shaping or introducing additional information to generate supervision signals of intermediate paths. But they required extra expert experience and label information. To address this situation, we propose a novel method under the teacher-student framework, it leverages perturbation consistency to learn intermediate paths. In the teacher network, we construct close data points for intermediate path prediction by applying random perturbations. Inspired by the data smoothing assumption that labels of close data points should be the same, a consistency loss over predictions of constructed data points and original ones is evaluated. The student network is used to answer questions more precisely by leveraging the intermediate distribution learned from the teacher network. Extensive experiments on two benchmark datasets are conducted, and the results have demonstrated the effectiveness of the proposed method.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76773025","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00179
Xue Li, Junjie Zhang, Junlong Ma
Multiple Choice Question Answering(MCQA) aims to automatically choose a correct answer from candidate options when given a passage and question. Existing approaches generally model attention mechanisms based on whole-passage information or manually tag key sentences for weakly supervised learning, which leads to the models focusing extensively on redundant information and costly manual annotation. In this paper, we consider evidence sentence extraction work in an unsupervised way to precisely pinpoint evidence sentences and minimize the impact of redundant information while avoiding costly manual annotations. Specifically, we propose a novel model called Term Similarity-aware Extensive and Intensive Reading(TS-EIR), which dynamically and automatically refines critical information by term similarity. In detail, it intelligently selects sentences more relevant to the question from the passage and deeply extracts features by enhanced graph convolutional neural network. We apply the proposed TS-EIR to a typical pre-trained language model, BERT, for encoding and evaluate it on the RACE and Dream benchmarks, which verify our model achieves substantial performance improvements over the current baseline.
{"title":"Term Similarity-aware Extensive and Intensive Reading For Multiple Choice Question Answering","authors":"Xue Li, Junjie Zhang, Junlong Ma","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00179","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00179","url":null,"abstract":"Multiple Choice Question Answering(MCQA) aims to automatically choose a correct answer from candidate options when given a passage and question. Existing approaches generally model attention mechanisms based on whole-passage information or manually tag key sentences for weakly supervised learning, which leads to the models focusing extensively on redundant information and costly manual annotation. In this paper, we consider evidence sentence extraction work in an unsupervised way to precisely pinpoint evidence sentences and minimize the impact of redundant information while avoiding costly manual annotations. Specifically, we propose a novel model called Term Similarity-aware Extensive and Intensive Reading(TS-EIR), which dynamically and automatically refines critical information by term similarity. In detail, it intelligently selects sentences more relevant to the question from the passage and deeply extracts features by enhanced graph convolutional neural network. We apply the proposed TS-EIR to a typical pre-trained language model, BERT, for encoding and evaluate it on the RACE and Dream benchmarks, which verify our model achieves substantial performance improvements over the current baseline.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76895686","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00077
Jingjie Wang, Xiang Wei, Siyang Lu, Mingquan Wang, Xiaoyu Liu, Wei Lu
Nowadays, the self-attention mechanism has become a resound of visual feature extraction along with convolution. The transformer network composed of self-attention has developed rapidly and made remarkable achievements in visual tasks. The self-attention shows the potential to replace convolution as the primary method of visual feature extraction in ubiquitous intelligence. Nevertheless, the development of the Visual Transformer still suffer from the following problems: a) The self-attention mechanism has a low inductive bias, which leads to large data demand and a high training cost. b) The Transformer backbone network cannot adapt well to the low visual information density and performs unsatisfactorily under low resolution and small-scale datasets. To tackle the abovementioned two problems, this paper proposes a novel algorithm based on the mature Visual Transformer architecture, which is dedicated to exploring the performance potential of the Transformer network and its kernel self-attention mechanism on small-scale datasets. Specifically, we first propose a network architecture equipped with multi-coordination strategy to solve the self-attention degradation problem inherent in the existing Transformer architecture. Secondly, we introduce consistent regularization into the Transformer to make the self-attention mechanism acquire more reliable feature representation ability in the case of insufficient visual features. In the experiments, CSwin Transformer, the mainstream visual model, is selected to verify the effectiveness of the proposed method on the prevalent small datasets, and superior results are achieved. In particular, without pre-training, our accuracy on the CIFAR-100 dataset is improved by 1.24% compared to CSwin.
{"title":"Redesign Visual Transformer For Small Datasets","authors":"Jingjie Wang, Xiang Wei, Siyang Lu, Mingquan Wang, Xiaoyu Liu, Wei Lu","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00077","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00077","url":null,"abstract":"Nowadays, the self-attention mechanism has become a resound of visual feature extraction along with convolution. The transformer network composed of self-attention has developed rapidly and made remarkable achievements in visual tasks. The self-attention shows the potential to replace convolution as the primary method of visual feature extraction in ubiquitous intelligence. Nevertheless, the development of the Visual Transformer still suffer from the following problems: a) The self-attention mechanism has a low inductive bias, which leads to large data demand and a high training cost. b) The Transformer backbone network cannot adapt well to the low visual information density and performs unsatisfactorily under low resolution and small-scale datasets. To tackle the abovementioned two problems, this paper proposes a novel algorithm based on the mature Visual Transformer architecture, which is dedicated to exploring the performance potential of the Transformer network and its kernel self-attention mechanism on small-scale datasets. Specifically, we first propose a network architecture equipped with multi-coordination strategy to solve the self-attention degradation problem inherent in the existing Transformer architecture. Secondly, we introduce consistent regularization into the Transformer to make the self-attention mechanism acquire more reliable feature representation ability in the case of insufficient visual features. In the experiments, CSwin Transformer, the mainstream visual model, is selected to verify the effectiveness of the proposed method on the prevalent small datasets, and superior results are achieved. In particular, without pre-training, our accuracy on the CIFAR-100 dataset is improved by 1.24% compared to CSwin.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80934160","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00240
Shaopeng Wang, Chunkai Feng
Since its introduction over five years ago, time series chain has become a fundamental tool for time series analytics, finding diverse uses in dozens of domains. Recent work has generalized the definition of time series chain, and introduced a novel definition of time series chain with directionality and graduality characteristics (TSC-DG) which can significantly enhance both robustness and usability of the original time series chain. However, previous studies on TSCDG process fixed-length time series. In this work, we focus on the issue of all-chain set with direction and graduality characteristics (all-TSCS-DG) mining over streaming time series for the first time, where all-TSCS-DG is the core of current TSCDG researches. We propose an improved Naive algorithm (IN) to solve this problem. Compared to the Naive, the IN guarantees the same space costs and results firstly, secondly is the IN takes two additional optimal strategies to further improve the time efficiency. The basic ideas of these two strategies are both incremental computing. The first one can make the IN update the IB structure at each time-tick incrementally, where the IB is an important data structure that is used to obtain the all-TSCS-DG. The second one makes the IN obtain mining results at current time-tick based on the ones at the last time-tick incrementally. Extensive experiments on real dataset demonstrate the efficiency and effectiveness of the IN.
{"title":"Discovering All-chain Set with Direction and Graduality Characteristics over Streaming Time Series","authors":"Shaopeng Wang, Chunkai Feng","doi":"10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00240","DOIUrl":"https://doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00240","url":null,"abstract":"Since its introduction over five years ago, time series chain has become a fundamental tool for time series analytics, finding diverse uses in dozens of domains. Recent work has generalized the definition of time series chain, and introduced a novel definition of time series chain with directionality and graduality characteristics (TSC-DG) which can significantly enhance both robustness and usability of the original time series chain. However, previous studies on TSCDG process fixed-length time series. In this work, we focus on the issue of all-chain set with direction and graduality characteristics (all-TSCS-DG) mining over streaming time series for the first time, where all-TSCS-DG is the core of current TSCDG researches. We propose an improved Naive algorithm (IN) to solve this problem. Compared to the Naive, the IN guarantees the same space costs and results firstly, secondly is the IN takes two additional optimal strategies to further improve the time efficiency. The basic ideas of these two strategies are both incremental computing. The first one can make the IN update the IB structure at each time-tick incrementally, where the IB is an important data structure that is used to obtain the all-TSCS-DG. The second one makes the IN obtain mining results at current time-tick based on the ones at the last time-tick incrementally. Extensive experiments on real dataset demonstrate the efficiency and effectiveness of the IN.","PeriodicalId":43791,"journal":{"name":"Scalable Computing-Practice and Experience","volume":null,"pages":null},"PeriodicalIF":1.1,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79525821","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","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：