Neural Machine Translation (NMT) models based on Transformer achieve promising progress in both translation quality and training speed. Such a strong framework adopts parallel structures that greatly improve the decoding speed without losing quality. However, due to the self-attention network in decoder that cannot maintain the parallelization under the auto-regressive scheme, the Transformer did not enjoy the same speed performance as training when inference. In this work, with simplicity and feasibility in mind, we introduce a gated cumulative attention network to replace the self-attention part in Transformer decoder to maintain the parallelization property in the inference phase. The gated cumulative attention network includes two sub-layers, a gated linearly cumulative layer that creates the relationship between already predicted tokens and current representation, and a feature fusion layer that enhances the representation with a feature fusion operation. The proposed method was evaluated on WMT17 datasets with 12 language pair groups. Experimental results show the effectiveness of the proposed method and also demonstrated that the proposed gated cumulative attention network has adequate ability as an alternative to the self-attention part in the Transformer decoder.
{"title":"Accelerating Transformer for Neural Machine Translation","authors":"Li Huang, Wenyu Chen, Hong Qu","doi":"10.1145/3457682.3457711","DOIUrl":"https://doi.org/10.1145/3457682.3457711","url":null,"abstract":"Neural Machine Translation (NMT) models based on Transformer achieve promising progress in both translation quality and training speed. Such a strong framework adopts parallel structures that greatly improve the decoding speed without losing quality. However, due to the self-attention network in decoder that cannot maintain the parallelization under the auto-regressive scheme, the Transformer did not enjoy the same speed performance as training when inference. In this work, with simplicity and feasibility in mind, we introduce a gated cumulative attention network to replace the self-attention part in Transformer decoder to maintain the parallelization property in the inference phase. The gated cumulative attention network includes two sub-layers, a gated linearly cumulative layer that creates the relationship between already predicted tokens and current representation, and a feature fusion layer that enhances the representation with a feature fusion operation. The proposed method was evaluated on WMT17 datasets with 12 language pair groups. Experimental results show the effectiveness of the proposed method and also demonstrated that the proposed gated cumulative attention network has adequate ability as an alternative to the self-attention part in the Transformer decoder.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129134509","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}
Wenhan Deng, Ming Zhang, Kai He, Lijin Wang, Juan Lin, Yiwen Zhong
Closed Loop Layout Problem (CLLP) is an NP-hard facility layout problem that determines the most favorable placement of facilities along a rectangle loop with adjustable size. The primary objective of the CLLP is to minimize the total transportation cost of the material flow between facilities. To obtain this objective, the optimal placement sequence of the facilities and the corresponding optimal size of the rectangle loop must be obtained at the same time. Although several metaheuristic-based methods have been proposed to tackle the CLLP, those methods only use metaheuristics to search the optimal placement sequence of facilities, and the optimal size of the rectangle loop is obtained by enumeration method. In order to improve the search efficiency of metaheuristics for the CLLP, this paper presents a Mixed-coding Harmony Search (MHS) algorithm which includes Permutation-based Discrete Harmony Search (PDHS) and Continuous Harmony Search (CHS). The PDHS part is designed to search the optimal placement sequence of facilities, and the CHS part is used to find the optimal size of the rectangle loop. Comparing experiments, which were conducted on 13 CLLP instances, have shown that the MHS algorithm obtains better results in less time than other existing metaheuristics.
{"title":"A Mixed-coding Harmony Search Algorithm for the Closed Loop Layout Problem","authors":"Wenhan Deng, Ming Zhang, Kai He, Lijin Wang, Juan Lin, Yiwen Zhong","doi":"10.1145/3457682.3457696","DOIUrl":"https://doi.org/10.1145/3457682.3457696","url":null,"abstract":"Closed Loop Layout Problem (CLLP) is an NP-hard facility layout problem that determines the most favorable placement of facilities along a rectangle loop with adjustable size. The primary objective of the CLLP is to minimize the total transportation cost of the material flow between facilities. To obtain this objective, the optimal placement sequence of the facilities and the corresponding optimal size of the rectangle loop must be obtained at the same time. Although several metaheuristic-based methods have been proposed to tackle the CLLP, those methods only use metaheuristics to search the optimal placement sequence of facilities, and the optimal size of the rectangle loop is obtained by enumeration method. In order to improve the search efficiency of metaheuristics for the CLLP, this paper presents a Mixed-coding Harmony Search (MHS) algorithm which includes Permutation-based Discrete Harmony Search (PDHS) and Continuous Harmony Search (CHS). The PDHS part is designed to search the optimal placement sequence of facilities, and the CHS part is used to find the optimal size of the rectangle loop. Comparing experiments, which were conducted on 13 CLLP instances, have shown that the MHS algorithm obtains better results in less time than other existing metaheuristics.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129213362","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 traditional artificial neural network based on gradient descent method result in low computational efficiency and local convergence for transient electromagnetic inversion. To solve the these problems, a hybrid approach combining principal component analysis (PCA) and online sequential extreme learning machine (OSELM) is proposed in this paper (PCA-OSELM) and is applied in the transient electromagnetic inversion. First, a principal component analysis method is introduced to reduce the dimension of vertical magnetic field data and improves the computational efficiency. Then, the new samples obtained from the data sets are added to the training samples as the next update information to establish the OSELM prediction models, so that improve the inversion accuracy. Finally, the inversion results of the two typical layered geoelectric models and a quasi two-dimensional geoelectric model show that the proposed approach can well solve the modeling nonlinear problem that high-dimensional data generated by transient electromagnetic method. Moreover, compared with other nonlinear inversion methods (OSELM, ELM), the PCA-OSELM achieves more accurate, better generalization ability and higher computational efficiency, which can provide new ideas for the application of neural networks in geophysical inversion.
{"title":"Using Principal Component Analysis and Online Sequential Extreme Learning Machine Approach for Transient Electromagnetic Nonlinear Inversion: TEM-Inversion-based-on-PCA-OSELM","authors":"Ruiyou Li","doi":"10.1145/3457682.3457766","DOIUrl":"https://doi.org/10.1145/3457682.3457766","url":null,"abstract":"The traditional artificial neural network based on gradient descent method result in low computational efficiency and local convergence for transient electromagnetic inversion. To solve the these problems, a hybrid approach combining principal component analysis (PCA) and online sequential extreme learning machine (OSELM) is proposed in this paper (PCA-OSELM) and is applied in the transient electromagnetic inversion. First, a principal component analysis method is introduced to reduce the dimension of vertical magnetic field data and improves the computational efficiency. Then, the new samples obtained from the data sets are added to the training samples as the next update information to establish the OSELM prediction models, so that improve the inversion accuracy. Finally, the inversion results of the two typical layered geoelectric models and a quasi two-dimensional geoelectric model show that the proposed approach can well solve the modeling nonlinear problem that high-dimensional data generated by transient electromagnetic method. Moreover, compared with other nonlinear inversion methods (OSELM, ELM), the PCA-OSELM achieves more accurate, better generalization ability and higher computational efficiency, which can provide new ideas for the application of neural networks in geophysical inversion.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126770743","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}
In this project, we derived an effective and efficient mathematical algorithm to identify bird species based on bird calls. Classifying bird species can be useful in real applications, such as determining the health of an ecosystem, or identifying hazardous species of birds near airports and reducing the bird-aircraft strikes. Having well-trained ornithologists to identify the characteristics of birds requires many man hours, and the results may be subjective. Our research was intended to develop a semi-automatic classification algorithm. We first performed a wavelet decomposition algorithm over more than 1200 syllables from 12 different bird species, and then extracted a set of eight parameters from each instance. The dataset formed by the instances and associated parameters was used to train and test different classifiers. Our results showed that among all the classifiers we tested, Cubic Support Vector Machine and Random Forest achieved the highest classification rates, each of which was over 93%.
{"title":"Acoustic Classification of Bird Species Using Wavelets and Learning Algorithms","authors":"Song Yang, R. Frier, Qiang Shi","doi":"10.1145/3457682.3457692","DOIUrl":"https://doi.org/10.1145/3457682.3457692","url":null,"abstract":"In this project, we derived an effective and efficient mathematical algorithm to identify bird species based on bird calls. Classifying bird species can be useful in real applications, such as determining the health of an ecosystem, or identifying hazardous species of birds near airports and reducing the bird-aircraft strikes. Having well-trained ornithologists to identify the characteristics of birds requires many man hours, and the results may be subjective. Our research was intended to develop a semi-automatic classification algorithm. We first performed a wavelet decomposition algorithm over more than 1200 syllables from 12 different bird species, and then extracted a set of eight parameters from each instance. The dataset formed by the instances and associated parameters was used to train and test different classifiers. Our results showed that among all the classifiers we tested, Cubic Support Vector Machine and Random Forest achieved the highest classification rates, each of which was over 93%.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125266993","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}
M. Salampasis, Theodosios Siomos, Alkiviadis Katsalis, K. Diamantaras, Konstantinos Christantonis, Marina Delianidi, Iphigenia Karaveli
Recurrent Neural Networks (RNNs) have been shown to perform very effectively in session-based recommendation settings, when compared to other commonly used methods that consider the last viewed item of the user and precomputed item-to-item similarities. However, there is little systematic study on how RNNs perform in comparison to methods that use embeddings for item representation for Collaborative Filtering. In this paper we use two industry datasets to compare RNNs with other sequential recommenders that use various embedding methods to represent items. The first dataset corresponds to a typical e-commerce session-based scenario demanding effective next-item recommendation. The second dataset represents a last-basket prediction setting. Results show that although the RNN greatly outperforms embedding methods in the next-item scenario, the dynamic nature of the RNNs has not the same positive impact in the last-basket prediction task. We also present and test a framework that enables the hybrid utilization of text content and item sequences using embeddings. Finally, we report on experiments with reranking methods that demonstrate the effectiveness of simple and practical methods, using item categories, to improve the results.
{"title":"Comparison of RNN and Embeddings Methods for Next-item and Last-basket Session-based Recommendations","authors":"M. Salampasis, Theodosios Siomos, Alkiviadis Katsalis, K. Diamantaras, Konstantinos Christantonis, Marina Delianidi, Iphigenia Karaveli","doi":"10.1145/3457682.3457755","DOIUrl":"https://doi.org/10.1145/3457682.3457755","url":null,"abstract":"Recurrent Neural Networks (RNNs) have been shown to perform very effectively in session-based recommendation settings, when compared to other commonly used methods that consider the last viewed item of the user and precomputed item-to-item similarities. However, there is little systematic study on how RNNs perform in comparison to methods that use embeddings for item representation for Collaborative Filtering. In this paper we use two industry datasets to compare RNNs with other sequential recommenders that use various embedding methods to represent items. The first dataset corresponds to a typical e-commerce session-based scenario demanding effective next-item recommendation. The second dataset represents a last-basket prediction setting. Results show that although the RNN greatly outperforms embedding methods in the next-item scenario, the dynamic nature of the RNNs has not the same positive impact in the last-basket prediction task. We also present and test a framework that enables the hybrid utilization of text content and item sequences using embeddings. Finally, we report on experiments with reranking methods that demonstrate the effectiveness of simple and practical methods, using item categories, to improve the results.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"108 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122418354","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}
Xiaoshun Lu, Si Chen, Zhuoyuan Zheng, Chenyu Weng, Rui Xu
The traditional weighted multiple instance learning based online object tracking methods often use the Euclidean distance between the centers of the bounding boxes of the target and the instance to weight the instances in the positive bag, which can not effectively measure the contribution degree of the instances of the positive and negative bags and easily causes the object drifting problem. This paper proposes a generalized intersection over union based online weighted multiple instance learning algorithm (named GIoU-WMIL) for object tracking. This algorithm introduces a novel generalized intersection over union (GIoU) to calculate the overlap degree between the bounding boxes of the target and each instance in the bags, in order to effectively measure the contribution of the different instances. Furthermore, a new objective function is designed by employing the GIoU-based weights of all the instances in the positive and negative bags. Experiments show that the proposed algorithm has the good robustness and accuracy on several challenging video sequences.
{"title":"Generalized Intersection over Union Based Online Weighted Multiple Instance Learning Algorithm for Object Tracking","authors":"Xiaoshun Lu, Si Chen, Zhuoyuan Zheng, Chenyu Weng, Rui Xu","doi":"10.1145/3457682.3457698","DOIUrl":"https://doi.org/10.1145/3457682.3457698","url":null,"abstract":"The traditional weighted multiple instance learning based online object tracking methods often use the Euclidean distance between the centers of the bounding boxes of the target and the instance to weight the instances in the positive bag, which can not effectively measure the contribution degree of the instances of the positive and negative bags and easily causes the object drifting problem. This paper proposes a generalized intersection over union based online weighted multiple instance learning algorithm (named GIoU-WMIL) for object tracking. This algorithm introduces a novel generalized intersection over union (GIoU) to calculate the overlap degree between the bounding boxes of the target and each instance in the bags, in order to effectively measure the contribution of the different instances. Furthermore, a new objective function is designed by employing the GIoU-based weights of all the instances in the positive and negative bags. Experiments show that the proposed algorithm has the good robustness and accuracy on several challenging video sequences.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"86 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133358090","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}
Most methods to recognize the structure of a table are to use the object detection approach to directly locate each cell in the table or to segment the table line based on the fully convolutional network (FCN). The problem of the former is that it is laborious to recognize the distorted table, while the problem of the latter is that the sample imbalance makes it difficult to train the model. In this paper, a gradient heatmap based table structure recognition method is proposed, by exploring the gradient heatmaps of the vertical lines and horizontal lines in the table. Specifically, the pixels of the vertical lines of the table are obtained according to the gradient heatmap, then the pixels of the horizontal lines are obtained using the same method, and finally the table structure is restored by using the connected domain search method. Compared with the Single Shot MultiBox Detector (SSD) and Faster RCNN that directly detects cells, our Average Precision (AP) value reached up to 99.5%, which is much higher than the above models. Additionally, we demonstrate that the AP values of the proposed models are reduced almost negligibly when the IoU threshold increased from 0.5 to 0.75, while the AP value of the fast RCNN and SSD model decreased significantly.
{"title":"A Gradient heatmap based Table Structure Recognition","authors":"Lingjun Kong, Yunchao Bao, Qianwen Wang, Lijun Cao, Shengmei Zhao","doi":"10.1145/3457682.3457752","DOIUrl":"https://doi.org/10.1145/3457682.3457752","url":null,"abstract":"Most methods to recognize the structure of a table are to use the object detection approach to directly locate each cell in the table or to segment the table line based on the fully convolutional network (FCN). The problem of the former is that it is laborious to recognize the distorted table, while the problem of the latter is that the sample imbalance makes it difficult to train the model. In this paper, a gradient heatmap based table structure recognition method is proposed, by exploring the gradient heatmaps of the vertical lines and horizontal lines in the table. Specifically, the pixels of the vertical lines of the table are obtained according to the gradient heatmap, then the pixels of the horizontal lines are obtained using the same method, and finally the table structure is restored by using the connected domain search method. Compared with the Single Shot MultiBox Detector (SSD) and Faster RCNN that directly detects cells, our Average Precision (AP) value reached up to 99.5%, which is much higher than the above models. Additionally, we demonstrate that the AP values of the proposed models are reduced almost negligibly when the IoU threshold increased from 0.5 to 0.75, while the AP value of the fast RCNN and SSD model decreased significantly.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"85 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125948020","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}
In the case of the training dataset suffering from heavy class-imbalance, deep learning algorithms may perform poorly. Due to the data-poor, the neural network cannot fully learn the representation of minority classes. In this paper, we proposed a data augmentation strategy called Intra-Class Cutmix for unbalanced datasets. Our algorithm can enhance the learning ability of neural networks for minority classes by mixing the intra-class samples of minority classes, and correct the decision boundary affected by unbalanced datasets. Although the method is simple, for unbalanced datasets, our method can be used as a supplement to traditional data augmentation methods (such as Randomerasing, Cutmix, etc.) to further enhance the performance of the network. In addition, Intra-Class Cutmix is also suitable for advanced re-balancing strategies. We conducted experiments on the CIFAR-10, CIFAR-100 and Fashion-MNIST datasets. Our results proved the effectiveness and universality of our method.
{"title":"Intra-Class Cutmix for Unbalanced Data Augmentation","authors":"Caidan Zhao, Yang Lei","doi":"10.1145/3457682.3457719","DOIUrl":"https://doi.org/10.1145/3457682.3457719","url":null,"abstract":"In the case of the training dataset suffering from heavy class-imbalance, deep learning algorithms may perform poorly. Due to the data-poor, the neural network cannot fully learn the representation of minority classes. In this paper, we proposed a data augmentation strategy called Intra-Class Cutmix for unbalanced datasets. Our algorithm can enhance the learning ability of neural networks for minority classes by mixing the intra-class samples of minority classes, and correct the decision boundary affected by unbalanced datasets. Although the method is simple, for unbalanced datasets, our method can be used as a supplement to traditional data augmentation methods (such as Randomerasing, Cutmix, etc.) to further enhance the performance of the network. In addition, Intra-Class Cutmix is also suitable for advanced re-balancing strategies. We conducted experiments on the CIFAR-10, CIFAR-100 and Fashion-MNIST datasets. Our results proved the effectiveness and universality of our method.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122471757","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 general focus of this paper is the improvement of state-of-the-art Bayesian tracking filters specialized to the domain of ground moving target tracking to obtain high-quality track information by incorporation of road-map information into a Gaussian mixture probability hypothesis density (GM-PHD) filtering scheme. In this paper, we propose a road-labeled GM-PHD (GM-RL-PHD) filter for ground targets with road-segment constrained dynamics and the recursive equations of the filter is derived. The proposed filter is validated with a ground target tracking example. The simulation results show that the proposed algorithm can improve the performance of ground target tracking algorithm by fusing road map information.
{"title":"Tracking Ground Targets with Road Constraint Using a Gaussian Mixture Road-Labeled PHD Filter","authors":"Jihong Zheng, Jinming Min, He He","doi":"10.1145/3457682.3457738","DOIUrl":"https://doi.org/10.1145/3457682.3457738","url":null,"abstract":"The general focus of this paper is the improvement of state-of-the-art Bayesian tracking filters specialized to the domain of ground moving target tracking to obtain high-quality track information by incorporation of road-map information into a Gaussian mixture probability hypothesis density (GM-PHD) filtering scheme. In this paper, we propose a road-labeled GM-PHD (GM-RL-PHD) filter for ground targets with road-segment constrained dynamics and the recursive equations of the filter is derived. The proposed filter is validated with a ground target tracking example. The simulation results show that the proposed algorithm can improve the performance of ground target tracking algorithm by fusing road map information.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124155789","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}
We put forward a powerful facial expression recognition method based on removing the noise features from the input image to achieve a significant improvement in accuracy. To achieve this goal, we first exploit GAN network to generate a neutral face from the emotional face, and then apply two different convolution layers to extract emotional face features and neutral face features separately. Finally, we eliminate neutral face features from emotional face features to get pure “emotion features”, which are then used to get prediction results. The overall prediction network, termed GAN Erased Facial Feature Network (GE2FN) achieves 98.02% ACC on the CK+ dataset with 48x48 input. The accuracy rate is significantly improved compared to other approaches, including the current mainstream VGG approach
{"title":"An Efficacious Method for Facial Expression Recognition: GAN Erased Facial Feature Network (GE2FN)","authors":"Tao Zhang, Tang Kai","doi":"10.1145/3457682.3457746","DOIUrl":"https://doi.org/10.1145/3457682.3457746","url":null,"abstract":"We put forward a powerful facial expression recognition method based on removing the noise features from the input image to achieve a significant improvement in accuracy. To achieve this goal, we first exploit GAN network to generate a neutral face from the emotional face, and then apply two different convolution layers to extract emotional face features and neutral face features separately. Finally, we eliminate neutral face features from emotional face features to get pure “emotion features”, which are then used to get prediction results. The overall prediction network, termed GAN Erased Facial Feature Network (GE2FN) achieves 98.02% ACC on the CK+ dataset with 48x48 input. The accuracy rate is significantly improved compared to other approaches, including the current mainstream VGG approach","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"60 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134287670","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：