Recently, attention based sequence-to-sequence model become a research hotspot in speech recognition. The attention model has the problem of slow convergence and poor robustness. In this paper, a model that jointed a bottleneck feature extraction network and attention model is proposed. The model is composed of a Deep Belief Network as bottleneck feature extraction network and an attention-based encoder-decoder model. DBN can store the priori information from Hidden Markov Model so that increasing convergence speed of and enhancing both robustness and discrimination of features. Attention model utilizes the temporal information of feature sequence to calculate the posterior probability of phoneme. Then the number of stack recurrent neural network layers in attention model is reduced in order to decrease the calculation of gradient. Experiments in the TIMIT corpus showed that the phoneme error rate is 17.80% in test set, the average training iteration decreased 52%, and the number of training iterations decreased from 139 to 89. The word error rate of WSJ eval92 is 12.9% without any external language model.
{"title":"Joint bottleneck feature and attention model for speech recognition","authors":"Long Xingyan, Qu Dan","doi":"10.1145/3208788.3208798","DOIUrl":"https://doi.org/10.1145/3208788.3208798","url":null,"abstract":"Recently, attention based sequence-to-sequence model become a research hotspot in speech recognition. The attention model has the problem of slow convergence and poor robustness. In this paper, a model that jointed a bottleneck feature extraction network and attention model is proposed. The model is composed of a Deep Belief Network as bottleneck feature extraction network and an attention-based encoder-decoder model. DBN can store the priori information from Hidden Markov Model so that increasing convergence speed of and enhancing both robustness and discrimination of features. Attention model utilizes the temporal information of feature sequence to calculate the posterior probability of phoneme. Then the number of stack recurrent neural network layers in attention model is reduced in order to decrease the calculation of gradient. Experiments in the TIMIT corpus showed that the phoneme error rate is 17.80% in test set, the average training iteration decreased 52%, and the number of training iterations decreased from 139 to 89. The word error rate of WSJ eval92 is 12.9% without any external language model.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"44 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121004132","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}
Xing Wang, Yixing Sun, Xiaoliang Tang, Ji Chen, Jiuxiang Jin
There is much fuzzy and non-monotonic knowledge in the Semantic Web Criminal Law Area. In recent years, the problem of fuzzy rules interchange has become one of the most important problems in the Semantic Web. Aiming at the problem of heterogeneous fuzzy rule interchange in the Semantic Web, which is based on the proposed Semantic Web Criminal Law Rule Language (CLRL), and based on the rules and norms of XML. We construct the rule mapping between the CLRL and Legal Knowledge Interchange Format (LKIF), and propose a heterogeneous fuzzy criminal law rules interchange architecture (CRIAXS), which supports the bidirectional rule interchange between legal rules. We also analyze the problem of information loss caused by the different language expression ability in the process of legal knowledge interchange, and put forward to the solution. Based on the above description, the prototype system CRIAXS which is based on the JavaScript language has been achieved on the HBuilder platform. We also verify the correctness and stability of the system through multiple conversion examples. The results show that the architecture and the implemented system which laysa solid foundation for the rule-based reasoning, has a good solution to the communication problem between heterogeneous systems, and it has a wide range of application prospects.
{"title":"Interchange of criminal rules between CLRL and LKIF","authors":"Xing Wang, Yixing Sun, Xiaoliang Tang, Ji Chen, Jiuxiang Jin","doi":"10.1145/3208788.3208802","DOIUrl":"https://doi.org/10.1145/3208788.3208802","url":null,"abstract":"There is much fuzzy and non-monotonic knowledge in the Semantic Web Criminal Law Area. In recent years, the problem of fuzzy rules interchange has become one of the most important problems in the Semantic Web. Aiming at the problem of heterogeneous fuzzy rule interchange in the Semantic Web, which is based on the proposed Semantic Web Criminal Law Rule Language (CLRL), and based on the rules and norms of XML. We construct the rule mapping between the CLRL and Legal Knowledge Interchange Format (LKIF), and propose a heterogeneous fuzzy criminal law rules interchange architecture (CRIAXS), which supports the bidirectional rule interchange between legal rules. We also analyze the problem of information loss caused by the different language expression ability in the process of legal knowledge interchange, and put forward to the solution. Based on the above description, the prototype system CRIAXS which is based on the JavaScript language has been achieved on the HBuilder platform. We also verify the correctness and stability of the system through multiple conversion examples. The results show that the architecture and the implemented system which laysa solid foundation for the rule-based reasoning, has a good solution to the communication problem between heterogeneous systems, and it has a wide range of application prospects.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128991213","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}
Hang Li, Zhuang Miao, Yang Li, Jiabao Wang, Yafei Zhang
To address the issue of background subtraction include shadow challenge, an online robust principal component analysis (RPCA) method with box constraint (BC-RPCA) has been proposed to detect moving object and accelerate the RPCA like method. First of all, the BC-RPCA method considers the input image sequences as low rank background, sparse foreground and moving shadow. Then the Augmented Lagrangian method is used to convert the box constraint into the objective function and rank-1 modification for thin SVD is also employed to accelerate the solver via alternating direction method of multipliers (ADMM). Finally, the experiments demonstrated the proposed method works effectively and has low computational complexity during real-time application.
{"title":"Background subtraction via online box constrained RPCA","authors":"Hang Li, Zhuang Miao, Yang Li, Jiabao Wang, Yafei Zhang","doi":"10.1145/3208788.3208797","DOIUrl":"https://doi.org/10.1145/3208788.3208797","url":null,"abstract":"To address the issue of background subtraction include shadow challenge, an online robust principal component analysis (RPCA) method with box constraint (BC-RPCA) has been proposed to detect moving object and accelerate the RPCA like method. First of all, the BC-RPCA method considers the input image sequences as low rank background, sparse foreground and moving shadow. Then the Augmented Lagrangian method is used to convert the box constraint into the objective function and rank-1 modification for thin SVD is also employed to accelerate the solver via alternating direction method of multipliers (ADMM). Finally, the experiments demonstrated the proposed method works effectively and has low computational complexity during real-time application.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"2018 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114698974","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}
Wenjun Quan, Qing Zhou, Hai Nan, Yanbin Chen, Ping Wang
Clustering is a common method for data analysis where a good clustering helps users to better understand the data. As for clustering quality measurement, the mainly used are some objective measures, while some researchers also paid attention to users' goals and they proposed methods to get users involved in clustering. However, a good clustering must meet the satisfaction of the users. Apart from these objective measures and users' goals, whether the clustering is easy to understand is also important for clustering quality measurement, especially in high-dimensional data clustering, if the data points in the final clusters are with high dimensions, it will hinder users' understanding of the clustering results. With all these concerns considered, we proposed an index of users' satisfaction with high-dimensional data clustering. According to this index, we further put forward a user-satisfaction-based clustering method to better serve users' satisfaction. We first developed an optimization model about users' satisfaction, then we used genetic algorithm to solve this model and obtained some high-quality clusterings, after reclustering of the clusterings obtained in previous steps, a few representative high-quality clusterings are provided for users to select. The experiment results suggest that our method is effective to provide some representative clusterings with the clustering quality, users' goals and the interpretability of clustering results being well considered.
{"title":"A user-satisfaction-based clustering method","authors":"Wenjun Quan, Qing Zhou, Hai Nan, Yanbin Chen, Ping Wang","doi":"10.1145/3208788.3208789","DOIUrl":"https://doi.org/10.1145/3208788.3208789","url":null,"abstract":"Clustering is a common method for data analysis where a good clustering helps users to better understand the data. As for clustering quality measurement, the mainly used are some objective measures, while some researchers also paid attention to users' goals and they proposed methods to get users involved in clustering. However, a good clustering must meet the satisfaction of the users. Apart from these objective measures and users' goals, whether the clustering is easy to understand is also important for clustering quality measurement, especially in high-dimensional data clustering, if the data points in the final clusters are with high dimensions, it will hinder users' understanding of the clustering results. With all these concerns considered, we proposed an index of users' satisfaction with high-dimensional data clustering. According to this index, we further put forward a user-satisfaction-based clustering method to better serve users' satisfaction. We first developed an optimization model about users' satisfaction, then we used genetic algorithm to solve this model and obtained some high-quality clusterings, after reclustering of the clusterings obtained in previous steps, a few representative high-quality clusterings are provided for users to select. The experiment results suggest that our method is effective to provide some representative clusterings with the clustering quality, users' goals and the interpretability of clustering results being well considered.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127352036","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 Merrifield-Simmons index of a graph is defined as the total number of the independent sets of the graph. This paper mainly discussed the Merrifield- Simmons index of two classes of lexicographic product graphs of Corona graphs P(m)n [H] and C(m)n[H], with the specific expressions are given.
{"title":"The Merrifield-Simmons index of two classes of lexicographic product graphs of corona graphs","authors":"Min Guo, Wen-wen Tian, N. Li","doi":"10.1145/3208788.3208790","DOIUrl":"https://doi.org/10.1145/3208788.3208790","url":null,"abstract":"The Merrifield-Simmons index of a graph is defined as the total number of the independent sets of the graph. This paper mainly discussed the Merrifield- Simmons index of two classes of lexicographic product graphs of Corona graphs P(m)n [H] and C(m)n[H], with the specific expressions are given.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115513676","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 extreme learning machine (ELM) is a novel single hidden layer feedforward neural network. Compared with traditional neural network algorithm, ELM has the advantages of fast learning speed and good generalization performance. However, there are still some shortages that restrict the further development of ELM, such as the perturbation and multicollinearity in the linear model. To the adverse effects caused by the perturbation and the multicollinearity, this paper proposes ELM based on condition number and variance decomposition ratio (CVELM) for regression, which separates the interference terms in the model by condition number and variance decomposition ratio, and then manipulate the interference items with weighted. Finally, the output layer weight is calculated by the least square method. The proposed algorithm can not only get good stability of the algorithm, but also reduce the impact on the non-interference items when dealing with the interference terms. The regression experiments on several datasets show that the proposed method owns a good generalization performance and stability.
{"title":"Extreme learning machine for regression based on condition number and variance decomposition ratio","authors":"Meiyi Li, Weibiao Cai, Qingshuai Sun","doi":"10.1145/3208788.3208794","DOIUrl":"https://doi.org/10.1145/3208788.3208794","url":null,"abstract":"The extreme learning machine (ELM) is a novel single hidden layer feedforward neural network. Compared with traditional neural network algorithm, ELM has the advantages of fast learning speed and good generalization performance. However, there are still some shortages that restrict the further development of ELM, such as the perturbation and multicollinearity in the linear model. To the adverse effects caused by the perturbation and the multicollinearity, this paper proposes ELM based on condition number and variance decomposition ratio (CVELM) for regression, which separates the interference terms in the model by condition number and variance decomposition ratio, and then manipulate the interference items with weighted. Finally, the output layer weight is calculated by the least square method. The proposed algorithm can not only get good stability of the algorithm, but also reduce the impact on the non-interference items when dealing with the interference terms. The regression experiments on several datasets show that the proposed method owns a good generalization performance and stability.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"118 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123246222","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}
Normality assumption of the random errors and the random effects is a routinely used technique in data analysis. However, this assumption might be unreasonable in many practical cases. In this paper the limitation is relaxed by assuming that the random error follows a reproductive dispersion model and the random effect is distributed as a skew-normal distribution, which is termed as a multivariate skew-normal reproductive dispersion random effects model. We propose a Bayesian procedure to simultaneously estimate the random effects and the unknown parameters on the basis of the Gibbs sampler and Metropolis-Hastings algorithm. In the end, the Framingham cholesterol data example is employed to demonstrate the preceding proposed Bayesian methodologies.
{"title":"Bayesian analysis for multivariate skew-normal reproductive dispersion random effects models","authors":"Yuanying Zhao, Xingde Duan, De-Wang Li","doi":"10.1145/3208788.3208805","DOIUrl":"https://doi.org/10.1145/3208788.3208805","url":null,"abstract":"Normality assumption of the random errors and the random effects is a routinely used technique in data analysis. However, this assumption might be unreasonable in many practical cases. In this paper the limitation is relaxed by assuming that the random error follows a reproductive dispersion model and the random effect is distributed as a skew-normal distribution, which is termed as a multivariate skew-normal reproductive dispersion random effects model. We propose a Bayesian procedure to simultaneously estimate the random effects and the unknown parameters on the basis of the Gibbs sampler and Metropolis-Hastings algorithm. In the end, the Framingham cholesterol data example is employed to demonstrate the preceding proposed Bayesian methodologies.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114711996","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}
Pests and diseases are a threat to paddy production, especially in Indonesia, but identification remains to be a challenge in massive scale and automatically. Increasing smartphone usage and deep learning advance create an opportunity to answer this problem. Collecting 4,511 images from four language using search engines, and augment it to develop diverse data set. This dataset fed into CaffeNet model and processed with Caffe framework. Experiment result in the model achieved accuracy 87%, which is higher than random selection 7.6%.
{"title":"Deep learning based classification for paddy pests & diseases recognition","authors":"Ahmad Arib Alfarisy, Quan Chen, M. Guo","doi":"10.1145/3208788.3208795","DOIUrl":"https://doi.org/10.1145/3208788.3208795","url":null,"abstract":"Pests and diseases are a threat to paddy production, especially in Indonesia, but identification remains to be a challenge in massive scale and automatically. Increasing smartphone usage and deep learning advance create an opportunity to answer this problem. Collecting 4,511 images from four language using search engines, and augment it to develop diverse data set. This dataset fed into CaffeNet model and processed with Caffe framework. Experiment result in the model achieved accuracy 87%, which is higher than random selection 7.6%.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124992075","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}
Now, Chinese economic development strategy is focusing on the restructuring of industrial structure, and the high-tech enterprises are facing great opportunities. However, due to the development and evaluation risks, investors are hard to assess their risks accurately. This paper proposed RSNCL-ANN ensemble strategies to build a risk assessment model and establishes indicators that cover corporate debt service, profitability, management, ownership structure and other aspects. These indicators are used to build a comprehensive and complete index system. In the RSNCL-ANN model, the neural network model was used as the base learner, and the strategies of random subspace and negative correlation learning were used to increase the diversity of the base learner so as to enhance the generalization ability of the integrated model. The experiment proved that this model had better predictive ability for venture firms.
{"title":"Credit risk assessment of high-tech enterprises based on RSNCL-ANN ensemble model","authors":"Maoguang Wang, Jiayu Yu, Zijian Ji","doi":"10.1145/3208788.3208801","DOIUrl":"https://doi.org/10.1145/3208788.3208801","url":null,"abstract":"Now, Chinese economic development strategy is focusing on the restructuring of industrial structure, and the high-tech enterprises are facing great opportunities. However, due to the development and evaluation risks, investors are hard to assess their risks accurately. This paper proposed RSNCL-ANN ensemble strategies to build a risk assessment model and establishes indicators that cover corporate debt service, profitability, management, ownership structure and other aspects. These indicators are used to build a comprehensive and complete index system. In the RSNCL-ANN model, the neural network model was used as the base learner, and the strategies of random subspace and negative correlation learning were used to increase the diversity of the base learner so as to enhance the generalization ability of the integrated model. The experiment proved that this model had better predictive ability for venture firms.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"56 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128276116","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 emergence of prescription fraud will reduce the effectiveness of health insurance investment. This paper will propose a new model to identify potentially fraudulent prescriptions and apply it to real prescription data to test its performance. Because of the low efficiency and high cost of prescription fraud through artificial experts, and because of the limitations of human knowledge, artificial detection is slow and insensitive to new fraud. We used the statistical characteristics of prescription data and other features related to the prescription to measure the risk level of the prescription, and found a prescription with high risk. The potential of this model can be used not only for off-line and online analysis and prediction of prescription fraud, but also for automatic updating of new fraud prescriptions. We test the model on real prescription data sets and compared to other approaches. The experimental results show that our model is promising for discovering the prescription fraud from the real health care data sets.
{"title":"Prescription fraud detection through statistic modeling","authors":"Hongxiang Zhang, Lizhen Wang","doi":"10.1145/3208788.3208803","DOIUrl":"https://doi.org/10.1145/3208788.3208803","url":null,"abstract":"The emergence of prescription fraud will reduce the effectiveness of health insurance investment. This paper will propose a new model to identify potentially fraudulent prescriptions and apply it to real prescription data to test its performance. Because of the low efficiency and high cost of prescription fraud through artificial experts, and because of the limitations of human knowledge, artificial detection is slow and insensitive to new fraud. We used the statistical characteristics of prescription data and other features related to the prescription to measure the risk level of the prescription, and found a prescription with high risk. The potential of this model can be used not only for off-line and online analysis and prediction of prescription fraud, but also for automatic updating of new fraud prescriptions. We test the model on real prescription data sets and compared to other approaches. The experimental results show that our model is promising for discovering the prescription fraud from the real health care data sets.","PeriodicalId":211585,"journal":{"name":"Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116992412","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：