Pub Date : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00086
Li Liu, Wujun Yang, Zhixian Chang
Aiming at the problem of high energy consumption and resource fragmentation caused by unbalanced multidimensional resource usage of servers in current cloud data centers, a virtual machine placement algorithm based on improved discrete differential evolution(IDDE) algorithm was proposed. According to the multi-dimensional resource requirements of virtual machines, the population initialization was used to improve the convergence speed of the algorithm, and the discrete differential mutation and crossover operations were used to ensure the diversity of the population. A multi-group elite selection strategy based on $varepsilon$ relaxation was proposed to select the optimal virtual machine placement scheme and enhance the global search ability of the algorithm. The simulation results show that compared with the other three algorithms such as the DE algorithm, the IDDE algorithm has a certain improvement effect in reducing energy consumption, improving resource utilization and reducing resource fragmentation.
{"title":"Multi-Objective Virtual Machine Placement Algorithm Based on Improved Discrete Differential Evolution","authors":"Li Liu, Wujun Yang, Zhixian Chang","doi":"10.1109/ICNLP58431.2023.00086","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00086","url":null,"abstract":"Aiming at the problem of high energy consumption and resource fragmentation caused by unbalanced multidimensional resource usage of servers in current cloud data centers, a virtual machine placement algorithm based on improved discrete differential evolution(IDDE) algorithm was proposed. According to the multi-dimensional resource requirements of virtual machines, the population initialization was used to improve the convergence speed of the algorithm, and the discrete differential mutation and crossover operations were used to ensure the diversity of the population. A multi-group elite selection strategy based on $varepsilon$ relaxation was proposed to select the optimal virtual machine placement scheme and enhance the global search ability of the algorithm. The simulation results show that compared with the other three algorithms such as the DE algorithm, the IDDE algorithm has a certain improvement effect in reducing energy consumption, improving resource utilization and reducing resource fragmentation.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79092184","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00065
Gerrit Schumann, Jorge Marx Gómez
Contradiction detection (CD) is a subfield of Natural Language Inference (NLI) that is relevant to many domains where contradictory statements in texts should be avoided (e.g., in financial or regulatory documents). With the advent of large annotated NLI datasets, there has been an increased focus on supervised deep-learning approaches in this research area. However, since this training data does not necessarily reflect the characteristic properties of the application data, unsupervised CD approaches are still relevant for certain domains or languages. In this paper, we therefore take up a recently published unsupervised NLI approach, reproduce parts of the proposed sentence transformations, extend it with various modifications, and evaluate it for the sole task of contradiction detection. The results show that under the exclusion of certain transformations types, an accuracy of 71.42 can be achieved on the SNLI test dataset.
{"title":"Unsupervised Contradiction Detection using Sentence Transformations","authors":"Gerrit Schumann, Jorge Marx Gómez","doi":"10.1109/ICNLP58431.2023.00065","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00065","url":null,"abstract":"Contradiction detection (CD) is a subfield of Natural Language Inference (NLI) that is relevant to many domains where contradictory statements in texts should be avoided (e.g., in financial or regulatory documents). With the advent of large annotated NLI datasets, there has been an increased focus on supervised deep-learning approaches in this research area. However, since this training data does not necessarily reflect the characteristic properties of the application data, unsupervised CD approaches are still relevant for certain domains or languages. In this paper, we therefore take up a recently published unsupervised NLI approach, reproduce parts of the proposed sentence transformations, extend it with various modifications, and evaluate it for the sole task of contradiction detection. The results show that under the exclusion of certain transformations types, an accuracy of 71.42 can be achieved on the SNLI test dataset.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79216189","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00027
Yinghua Li, Yue Liu, Y. Liu, Yangge Qiao, Jinglu He
At present, the image super-resolution method based on convolutional neural network has achieved a very high PSNR, but the high-frequency information obtained by using the mean square error as the loss function is not sufficient, and when the scale factor is large, the detail texture of the restored image is blurred, and it is not completely consistent with the human visual perception. Therefore, this paper proposes an image super-resolution algorithm based on GAN. We modify the residual block of the original SRGAN generator network into three modules: Edge-Reconstruction network, Low-Frequency feature (LF-feature) extraction module and Residual network. The Edge-Reconstruction network reconstructs the edge of SR image, and the LF-feature extraction module extracts the low-frequency information of the image. After that, the two parts of information are fused and transmitted to Residual network to extract the high-frequency information of the image, and then the SR image is reconstructed and enlarged. And use skip connection in the network to increase the network depth. The training results show that our network has better performance in both objective evaluation indicators and subjective vision. Even with a large-scale factor, our network can recover fine texture information.
{"title":"Multi-level Feature Extraction and Edge Reconstruction Fused Generative Adversarial Networks for Image Super Resolution","authors":"Yinghua Li, Yue Liu, Y. Liu, Yangge Qiao, Jinglu He","doi":"10.1109/ICNLP58431.2023.00027","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00027","url":null,"abstract":"At present, the image super-resolution method based on convolutional neural network has achieved a very high PSNR, but the high-frequency information obtained by using the mean square error as the loss function is not sufficient, and when the scale factor is large, the detail texture of the restored image is blurred, and it is not completely consistent with the human visual perception. Therefore, this paper proposes an image super-resolution algorithm based on GAN. We modify the residual block of the original SRGAN generator network into three modules: Edge-Reconstruction network, Low-Frequency feature (LF-feature) extraction module and Residual network. The Edge-Reconstruction network reconstructs the edge of SR image, and the LF-feature extraction module extracts the low-frequency information of the image. After that, the two parts of information are fused and transmitted to Residual network to extract the high-frequency information of the image, and then the SR image is reconstructed and enlarged. And use skip connection in the network to increase the network depth. The training results show that our network has better performance in both objective evaluation indicators and subjective vision. Even with a large-scale factor, our network can recover fine texture information.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84894210","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00054
Dan Du, Minghu Jiang
Research on adult cross-language speech perception suggests that adults face challenges perceiving segments distinctions that are not employed contrastively in their own language. Cross-language speech perceptual similarity has played a significant role in predicting and explaining L2 speech perception. It’s necessary to carry out a perceptual assimilation study on Mandarin which is an influencing language and has an increasing number of L2 learners around the world. The present study investigates the perception of Mandarin consonants by native Urdu speakers. To this purpose, 15 Urdu speakers were tested in an assimilation task in which they were asked to assimilate Mandarin consonants to their native language categories. The results show that native language interfere with the perception of non-native consonants in a certain extend that is in conformity to what PAM proposed. Furthermore, the results expand the application of PAM in Mandarin and provide a baseline for future relevant studies.
{"title":"Perceptual assimilation of Mandarin consonants in second language acquisition","authors":"Dan Du, Minghu Jiang","doi":"10.1109/ICNLP58431.2023.00054","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00054","url":null,"abstract":"Research on adult cross-language speech perception suggests that adults face challenges perceiving segments distinctions that are not employed contrastively in their own language. Cross-language speech perceptual similarity has played a significant role in predicting and explaining L2 speech perception. It’s necessary to carry out a perceptual assimilation study on Mandarin which is an influencing language and has an increasing number of L2 learners around the world. The present study investigates the perception of Mandarin consonants by native Urdu speakers. To this purpose, 15 Urdu speakers were tested in an assimilation task in which they were asked to assimilate Mandarin consonants to their native language categories. The results show that native language interfere with the perception of non-native consonants in a certain extend that is in conformity to what PAM proposed. Furthermore, the results expand the application of PAM in Mandarin and provide a baseline for future relevant studies.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79747055","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00080
Xiangyu Li, Yao Liu, Gaige Chen, Jiantao Chang
Intelligent fault diagnosis of rolling bearing is of great importance to improve the predictive maintenance ability of key assets in the context of industrial big data and smart manufacturing. Due to the usually high cost or infeasibility of obtaining data labels, large amount of data is unlabeled in practical industrial scenarios, which poses a challenge for conducting data-driven bearing fault diagnosis. In view of the characteristics of non-stationary and low signal-to-noise ratio of bearing vibration signals and the fact of lacking labeled samples but there exist lots of unlabeled samples, this paper proposes an intelligent diagnosis method for bearing faults based on variational mode decomposition (VMD) and improved self-training semi-supervised ensemble learning. Firstly, the original vibration signal is decomposed into several intrinsic mode functions using VMD, then correlation coefficient criterion is used to select the bearing fault feature bands to improve the signal-to-noise ratio, then time domain features are extracted, the labeled samples are expanded by the improved self-training semisupervised learning model, and finally the bearing fault diagnosis model is established based on ensemble learning by stacking method. Through the validation on two different experimental data sets, the proposed method was able to effectively extract the bearing fault feature information and improve the model accuracy by using unlabeled data compared with typical supervised learning models and other comparative models, which can meet the demand for intelligent diagnosis of bearing fault under the scenario of lacking labeled samples in real industries.
{"title":"Intelligent Fault Diagnosis of Rolling Bearing based on VMD and Improved Self-training Semi-supervised Ensemble Learning","authors":"Xiangyu Li, Yao Liu, Gaige Chen, Jiantao Chang","doi":"10.1109/ICNLP58431.2023.00080","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00080","url":null,"abstract":"Intelligent fault diagnosis of rolling bearing is of great importance to improve the predictive maintenance ability of key assets in the context of industrial big data and smart manufacturing. Due to the usually high cost or infeasibility of obtaining data labels, large amount of data is unlabeled in practical industrial scenarios, which poses a challenge for conducting data-driven bearing fault diagnosis. In view of the characteristics of non-stationary and low signal-to-noise ratio of bearing vibration signals and the fact of lacking labeled samples but there exist lots of unlabeled samples, this paper proposes an intelligent diagnosis method for bearing faults based on variational mode decomposition (VMD) and improved self-training semi-supervised ensemble learning. Firstly, the original vibration signal is decomposed into several intrinsic mode functions using VMD, then correlation coefficient criterion is used to select the bearing fault feature bands to improve the signal-to-noise ratio, then time domain features are extracted, the labeled samples are expanded by the improved self-training semisupervised learning model, and finally the bearing fault diagnosis model is established based on ensemble learning by stacking method. Through the validation on two different experimental data sets, the proposed method was able to effectively extract the bearing fault feature information and improve the model accuracy by using unlabeled data compared with typical supervised learning models and other comparative models, which can meet the demand for intelligent diagnosis of bearing fault under the scenario of lacking labeled samples in real industries.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84010040","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 : 2023-03-01DOI: 10.1109/icnlp58431.2023.00091
Liyuan Feng, Wujun Yang
With the rise of mobile edge computing, how to deal with the problem of edge computing task offloading has become one of the research hotspots. In order to solve the problem of serious congestion on wireless communication link caused by multi-users unloading to MEC server at the same time and competition for server computing resources among multi-user tasks after unloading, a joint optimization method for offloading decision and resource allocation was proposed. In this paper, a system task offloading model based on OFDMA technology is proposed, which takes into account the intensive and indivisible task resources generated by each user device. On this basis, a dynamic task offloading and resource allocation algorithm based on Nature DQN is proposed to solve the multi-client optimal offloading decision and multi-client computing resource allocation scheme. Finally, the simulation results show that the proposed task offloading model and the computational offloading algorithm based on Nature DQN are effective in optimizing the total delay of the long-term system.
{"title":"Multi-user Computing Offloading Based on Deep Reinforcement Learning","authors":"Liyuan Feng, Wujun Yang","doi":"10.1109/icnlp58431.2023.00091","DOIUrl":"https://doi.org/10.1109/icnlp58431.2023.00091","url":null,"abstract":"With the rise of mobile edge computing, how to deal with the problem of edge computing task offloading has become one of the research hotspots. In order to solve the problem of serious congestion on wireless communication link caused by multi-users unloading to MEC server at the same time and competition for server computing resources among multi-user tasks after unloading, a joint optimization method for offloading decision and resource allocation was proposed. In this paper, a system task offloading model based on OFDMA technology is proposed, which takes into account the intensive and indivisible task resources generated by each user device. On this basis, a dynamic task offloading and resource allocation algorithm based on Nature DQN is proposed to solve the multi-client optimal offloading decision and multi-client computing resource allocation scheme. Finally, the simulation results show that the proposed task offloading model and the computational offloading algorithm based on Nature DQN are effective in optimizing the total delay of the long-term system.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82492373","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00041
Meng Liang, Yao Shi
In the context of Natural Language Processing (NLP), Named Entity Recognition (NER) plays a crucial role in tasks like entity relationship extraction and knowledge graph construction. The accuracy of Chinese NER heavily relies on the representation of word embeddings. However, traditional word representation methods like word2vec suffer from word ambiguity and singular word vectors. Similarly, BERT-based word embeddings also exhibit anisotropy. To tackle these challenges, we propose a novel NER method that leverages BERT-whitening and dynamic fusion of BERT’s output from different layers. The dynamic fusion module calculates a weighted sum of BERT’s output across multiple layers, while the whitening module applies a whitening operation to eliminate the anisotropy of word embeddings. By integrating these modules, our model effectively captures the characteristics of input words, providing robust support for subsequent decoding. We evaluate our approach on the CLUENER2020 Chinese fine-grained named entity recognition dataset. Experimental results demonstrate that our method outperforms the traditional BERT-BiLSTM-CRF model without external resources and data expansion, leading to significant improvements in performance.
{"title":"Named Entity Recognition Method Based on BERT-whitening and Dynamic Fusion Model","authors":"Meng Liang, Yao Shi","doi":"10.1109/ICNLP58431.2023.00041","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00041","url":null,"abstract":"In the context of Natural Language Processing (NLP), Named Entity Recognition (NER) plays a crucial role in tasks like entity relationship extraction and knowledge graph construction. The accuracy of Chinese NER heavily relies on the representation of word embeddings. However, traditional word representation methods like word2vec suffer from word ambiguity and singular word vectors. Similarly, BERT-based word embeddings also exhibit anisotropy. To tackle these challenges, we propose a novel NER method that leverages BERT-whitening and dynamic fusion of BERT’s output from different layers. The dynamic fusion module calculates a weighted sum of BERT’s output across multiple layers, while the whitening module applies a whitening operation to eliminate the anisotropy of word embeddings. By integrating these modules, our model effectively captures the characteristics of input words, providing robust support for subsequent decoding. We evaluate our approach on the CLUENER2020 Chinese fine-grained named entity recognition dataset. Experimental results demonstrate that our method outperforms the traditional BERT-BiLSTM-CRF model without external resources and data expansion, leading to significant improvements in performance.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88991284","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 : 2023-03-01DOI: 10.1109/icnlp58431.2023.00088
Shihao Fei, Junxuan Wang, Fan Jiang, Yuan Ren, Senhu Zhou
In Satellite-Terrestrial Integrated Networks (STIN), from the perspective of increasing the capacity of the networks, the user experience, and the adaptability to high-speed motion occasions, a non-cooperative multi-service network selection scheme based on Q-learning and game theory (QRSG) is proposed. QRSG first obtains the multi-service network utility through the fuzzy process and uses it as the reward of Q-learning. The state of Q-learning includes the quality of service (QoS) and price attributes of the network currently connected by the user, as well as the situation of the user speed. The corresponding network selection strategy is the action of Q-learning. Then, the user predicts the payoff of the network selection strategy through a game algorithm to avoid access to an overloaded network. In addition, Binary Exponential Backoff Algorithm is introduced in QRSG to solve the problem of inaccurate throughput prediction in the scenario where multiple users concurrently switch to the same service node (SN). Simulations reveal that: 1) With QRSG, users with different speeds and QoS requirements can adaptively switch to the most suitable network. 2) Compared with the existing algorithms, QRSG can increase network throughput by more than 8% and reduce the total number of switching by about 60% in the case of a maximum loss of 1 to 2% of the system fairness.
在星地融合网络(STIN)中，从提高网络容量、用户体验和适应高速运动场合的角度出发，提出了一种基于q -学习和博弈论(QRSG)的非合作多业务网络选择方案。QRSG首先通过模糊过程得到多业务网络效用，并将其作为q学习的奖励。Q-learning的状态包括用户当前连接的网络的QoS (quality of service)和价格属性，以及用户的网速情况。相应的网络选择策略是Q-learning的动作。然后，用户通过博弈算法预测网络选择策略的收益，以避免进入过载的网络。此外，在QRSG中引入了二进制指数回退算法(Binary Exponential Backoff Algorithm)，解决了多个用户同时切换到同一业务节点(SN)的场景下吞吐量预测不准确的问题。仿真结果表明:1)使用QRSG，不同速率和QoS要求的用户可以自适应切换到最适合的网络。2)与现有算法相比，在系统公平性最大损失1 ~ 2%的情况下，QRSG可使网络吞吐量提高8%以上，总交换次数减少约60%。
{"title":"Research on Switching Strategy with Reinforcement Learning and Game Theory in Satellite-Terrestrial Integrated Networks","authors":"Shihao Fei, Junxuan Wang, Fan Jiang, Yuan Ren, Senhu Zhou","doi":"10.1109/icnlp58431.2023.00088","DOIUrl":"https://doi.org/10.1109/icnlp58431.2023.00088","url":null,"abstract":"In Satellite-Terrestrial Integrated Networks (STIN), from the perspective of increasing the capacity of the networks, the user experience, and the adaptability to high-speed motion occasions, a non-cooperative multi-service network selection scheme based on Q-learning and game theory (QRSG) is proposed. QRSG first obtains the multi-service network utility through the fuzzy process and uses it as the reward of Q-learning. The state of Q-learning includes the quality of service (QoS) and price attributes of the network currently connected by the user, as well as the situation of the user speed. The corresponding network selection strategy is the action of Q-learning. Then, the user predicts the payoff of the network selection strategy through a game algorithm to avoid access to an overloaded network. In addition, Binary Exponential Backoff Algorithm is introduced in QRSG to solve the problem of inaccurate throughput prediction in the scenario where multiple users concurrently switch to the same service node (SN). Simulations reveal that: 1) With QRSG, users with different speeds and QoS requirements can adaptively switch to the most suitable network. 2) Compared with the existing algorithms, QRSG can increase network throughput by more than 8% and reduce the total number of switching by about 60% in the case of a maximum loss of 1 to 2% of the system fairness.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88838960","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00013
Haibo Ge, Yu An, Wenhao He, Haodong Feng, Chaofeng Huang, Shuxian Wang
Siamese convolutional neural network, which is a classic framework for object tracking, has received extensive attention from the research community. The method uses a convolutional neural network to obtain target features and matches them with the search area features to achieve target tracking. Aiming at the problems that multi-layer features are difficult to extract effectively and network model parameters are complex, a target tracking algorithm (MA-SiamRPN++) with mixed attention mechanism is proposed based on SiamRPN++. Firstly, the channel attention mechanism is inserted into the backbone network, and then the output features of the channel attention network are fed into the spatial attention network, so as to improve the efficiency of feature extraction in different convolution layers by using mixed attention. At the same time, the deep cross -correlation network is used to better retain the feature information that is conducive to tracking and reduce the parameter complexity of the network to maintain the tracking speed. Finally, experiments on OTB100, VOT2016, and the long-term tracking dataset LaSOT show that the tracker proposed in this paper achieves higher accuracy and success rate than other state-of-the-art trackers.
{"title":"Target Tracking Algorithm Based on Mixed Attention and Siamese Network","authors":"Haibo Ge, Yu An, Wenhao He, Haodong Feng, Chaofeng Huang, Shuxian Wang","doi":"10.1109/ICNLP58431.2023.00013","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00013","url":null,"abstract":"Siamese convolutional neural network, which is a classic framework for object tracking, has received extensive attention from the research community. The method uses a convolutional neural network to obtain target features and matches them with the search area features to achieve target tracking. Aiming at the problems that multi-layer features are difficult to extract effectively and network model parameters are complex, a target tracking algorithm (MA-SiamRPN++) with mixed attention mechanism is proposed based on SiamRPN++. Firstly, the channel attention mechanism is inserted into the backbone network, and then the output features of the channel attention network are fed into the spatial attention network, so as to improve the efficiency of feature extraction in different convolution layers by using mixed attention. At the same time, the deep cross -correlation network is used to better retain the feature information that is conducive to tracking and reduce the parameter complexity of the network to maintain the tracking speed. Finally, experiments on OTB100, VOT2016, and the long-term tracking dataset LaSOT show that the tracker proposed in this paper achieves higher accuracy and success rate than other state-of-the-art trackers.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77061572","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00043
Jin Wu, Zhaoqi Zhang, Bo Zhao, Yu Wang
With their outstanding performance in natural language processing tasks such as machine translation and semantic recognition, deep neural networks have attracted great attention from both academia and industry. For more complex NLP tasks, people try to add more parameters to the network, expand more layers, input larger data samples, to produce a large model to solve the complex task. However, it is not the case that the deeper the layers, the better the parameters. There is a large amount of redundant information in the parameters, which not only contributes nothing to the results, but also increases the computational burden of the model and the storage burden of the hardware. Eliminating a small amount of redundant information often has no effect on the recognition rate of the model, but slightly improves it [1], so the neural network model needs to be compressed. Existing compression methods include model pruning, parameter quantization, tensor decomposition, knowledge distmation, etc. [2] In this paper, model pruning algorithm is selected to implement a result-oriented data-driven pruning algorithm by introducing the propagation characteristics and inter-layer correlation of neural networks and automatic decision making based on Feature Map information. Finally, the effectiveness of the result - oriented data - driven pruning algorithm is proved by comparative experiments.
{"title":"Data-Driven Pruning Algorithm Based on Result Orientation","authors":"Jin Wu, Zhaoqi Zhang, Bo Zhao, Yu Wang","doi":"10.1109/ICNLP58431.2023.00043","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00043","url":null,"abstract":"With their outstanding performance in natural language processing tasks such as machine translation and semantic recognition, deep neural networks have attracted great attention from both academia and industry. For more complex NLP tasks, people try to add more parameters to the network, expand more layers, input larger data samples, to produce a large model to solve the complex task. However, it is not the case that the deeper the layers, the better the parameters. There is a large amount of redundant information in the parameters, which not only contributes nothing to the results, but also increases the computational burden of the model and the storage burden of the hardware. Eliminating a small amount of redundant information often has no effect on the recognition rate of the model, but slightly improves it [1], so the neural network model needs to be compressed. Existing compression methods include model pruning, parameter quantization, tensor decomposition, knowledge distmation, etc. [2] In this paper, model pruning algorithm is selected to implement a result-oriented data-driven pruning algorithm by introducing the propagation characteristics and inter-layer correlation of neural networks and automatic decision making based on Feature Map information. Finally, the effectiveness of the result - oriented data - driven pruning algorithm is proved by comparative experiments.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88876216","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：