{"title":"Proper initialization of Hidden Markov models for industrial applications","authors":"Tingting Liu, J. Lemeire, Lixin Yang","doi":"10.1109/ChinaSIP.2014.6889291","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889291","url":null,"abstract":"Hidden Markov models (HMMs) are widely employed in the field of industrial applications such as machine maintenance. However, how to improve the effectiveness and efficiency of HMM-based approach is still an open question. The traditional HMMs learning method (e.g. the Baum-Welch algorithm) starts from an initial model with pre-defined topology and randomly-chosen parameters, and iteratively updates the model parameters until convergence. Thus, there is the risk of falling into local optima and low convergence speed because of wrongly defined number of hidden states and randomness of initial parameters. In this paper, we proposed a Segmentation and Clustering (SnC) based initialization method for the Baum-Welch algorithm to approximately estimate the number of hidden states and the model parameters for HMMs. The SnC approach was validated on both synthetic and real industrial data.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"417 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132636694","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}
{"title":"Multiclass feature selection algorithms base on R-SVM","authors":"Qifeng Xu, Xuegong Zhang","doi":"10.1109/ChinaSIP.2014.6889298","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889298","url":null,"abstract":"Feature selection is an important task in machine learning. Most existing feature selection methods were designed for two-class classification problems. Multiclass feature selection algorithm is less available. R-SVM or Recursive SVM is a SVM-based embedded feature selection algorithm proposed by Zhang et al[5]. It provides the function of recursive feature selection and outperforms another similar method SVM-RFE (SVM Recursive Feature Elimination) on noisy data and has become popular in bioinformatics. But both R-SVM and SVM-RFE support only binary classification. We extend R-SVM to multi-class classification and also implement the multiclass SVM-RFE method in the workflow of R-SVM. Both methods achieve good performance applied to commonly used bioinformatics datasets.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134213890","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}
{"title":"Hashing based feature aggregating for fast image copy retrieval","authors":"Lingyu Yan, H. Ling, Cong Liu, Xinyu Ou","doi":"10.1109/ChinaSIP.2014.6889281","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889281","url":null,"abstract":"Recently the methods based on visual words have become very popular in near- duplicate retrieval and content identification. However, obtaining the visual vocabulary by quantization is very time-consuming and unscalable to large databases. In this paper, we propose a fast feature aggregating method for image representation which uses machine learning based hashing to achieve fast feature aggregation. Since the machine learning based hashing effectively preserves neighborhood structure of data, it yields visual words with strong discriminability. Furthermore, the generated binary codes leads image representation building to be of low-complexity, making it efficient and scalable to large scale databases. The evaluation shows that our approach significantly outperforms state-of-the-art methods.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115191466","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}
{"title":"Exploit the scale of big data for data privacy: An efficient scheme based on distance-preserving artificial noise and secret matrix transform","authors":"Xiaohua Li, Zifan Zhang","doi":"10.1109/ChinaSIP.2014.6889293","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889293","url":null,"abstract":"In this paper we show that the extensive results in blind/non-blind channel identification developed within the community of signal processing in communications can play an important role in guaranteeing big data privacy. It is widely believed that the sheer scale of big data makes most conventional data privacy techniques ineffective for big data. In contrast to this pessimistic common belief, we propose a scheme that exploits the sheer scale to guarantee privacy. This scheme uses jointly artificial noise and secret matrix transform to scramble the source data. Desirable data utility can be supported because the noise and the transform preserve some important geometric properties of the source data. With a comprehensive privacy analysis, we use the blind/non-blind channel identification theories to show that the secret transform matrix and the source data can not be estimated from the scrambled data. The artificial noise and the sheer scale of big data are critical for this purpose. Simulations of collaborative filtering are conducted to demonstrate the proposed scheme.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"150 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115897322","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}
Zhiyue Lin, Yinglian Xiao, I. Hirano, P. Kahrilas, J. Pandolfino
{"title":"Automated deviation and modoling of the pressure-geometry relationship of esophageal body in impedance planimetry studies","authors":"Zhiyue Lin, Yinglian Xiao, I. Hirano, P. Kahrilas, J. Pandolfino","doi":"10.1109/ChinaSIP.2014.6889312","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889312","url":null,"abstract":"This study aimed to develop an automated method to investigate the pressure-geometry relationship of esophageal body in patients with eosinophilic esophagitis (EoE). Esophageal cross-sectional areas (CSAs) and intrabag pressure data were measured in 25 EoE patients with a functional luminal imaging probe (FLIP) and analyzed by a manual reading and by a customized MATLAB program with three functions: minimizing artifacts with a median filter; isolating contraction waves with a nonlinear pulse detector and deriving median values of a minimal pressure segment and corresponding narrowest esophageal CSAs during each distension; and modeling the distensibility of the esophageal body with a polynomial regression technique to derive the distension slope and distension plateau. This regression technique was also applied to manual readings for comparison. Such comparison shows that results of automated method highly correlated with manual reading (R≥0.84) and offered advantages in minimizing the influence of artifacts and inter-observer variability.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123130959","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}
{"title":"A topological feature constraint for nonrigid registration","authors":"Xiangbo Lin, Xin-Ning Wang","doi":"10.1109/ChinaSIP.2014.6889272","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889272","url":null,"abstract":"To satisfy the topology preservation requirement in deformable image registration, this research adopts a topological feature, the sign of the polygon area, as a regularization constraint. The reasonability and effectiveness of the proposed model are validated using synthetic image registration and inter-subject brain MRI image registration experiments.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122649599","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}
{"title":"Noisy training for deep neural networks","authors":"Xiangtao Meng, Chao Liu, Zhiyong Zhang, Dong Wang","doi":"10.1109/ChinaSIP.2014.6889193","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889193","url":null,"abstract":"Deep neural networks (DNN) have gained remarkable success in speech recognition, partially attributed to its flexibility in learning complex patterns of speech signals. This flexibility, however, may lead to serious over-fitting and hence miserable performance degradation in adverse environments such as those with high ambient noises. We propose a noisy training approach to tackle this problem: by injecting noises into the training speech intentionally and randomly, more generalizable DNN models can be learned. This `noise injection' technique has been well-known to the neural computation community, however there is little knowledge if it would work for the DNN model which involves a highly complex objective function. The experiments presented in this paper confirm that the original assumptions of the noise injection approach largely holds when learning deep structures, and the noisy training may provide substantial performance improvement for DNN-based speech recognition.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129218997","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}
Qing Shen, W. Liu, W. Cui, Siliang Wu, Yimin D. Zhang, M. Amin
{"title":"Group sparsity based wideband DOA estimation for co-prime arrays","authors":"Qing Shen, W. Liu, W. Cui, Siliang Wu, Yimin D. Zhang, M. Amin","doi":"10.1109/ChinaSIP.2014.6889242","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889242","url":null,"abstract":"A novel wideband direction-of-arrival (DOA) estimation method is proposed for co-prime arrays. After decomposing the wideband signals into different frequencies/subbands through a discrete Fourier transform or, more generally, a filter bank system, the increased degrees of freedom provided by co-prime arrays are fully exploited with a group sparsity based signal reconstruction method. Simulation results show that this novel method can distinguish much more sources than the number of physical sensors. Compared with the existing narrowband DOA estimation method for co-prime arrays, the proposed wideband method achieves a significant performance improvement.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129407523","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}
{"title":"A motion tracking system for hand activity assessment","authors":"Chia-Hsiung Chen, Y. Hu, R. Radwin","doi":"10.1109/ChinaSIP.2014.6889256","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889256","url":null,"abstract":"A video-based hand-motion detection and tracking system is developed for the purpose of automated hand activity level assessment in a factory workplace. While the basic approach follows the sequential Bayesian estimation principle, a unique verification step is proposed to validate the quality of estimates. An experimental simulated workplace assembly line workstation is developed. The proposed system shows satisfactory performance and robustness due to the novel verification step.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128275490","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}
Congying Zhang, Masayuki Suzuki, Gakuto Kurata, M. Nishimura, N. Minematsu
{"title":"Leveraging phonetic context dependent invariant structure for continuous speech recognition","authors":"Congying Zhang, Masayuki Suzuki, Gakuto Kurata, M. Nishimura, N. Minematsu","doi":"10.1109/ChinaSIP.2014.6889200","DOIUrl":"https://doi.org/10.1109/ChinaSIP.2014.6889200","url":null,"abstract":"Speech acoustics intrinsically vary due to linguistic and non-linguistic factors. The invariant structure extracted from a given utterance is one of the long-span acoustic representations, where acoustic variation caused by non-linguistic factors can be removed reasonably. It expresses spectral contrasts between acoustic events in an utterance. In previous studies, the invariant structure was leveraged in continuous speech recognition for reranking the N-best candidates hypothesized by a traditional automatic speech recognition (ASR) system. Use of the invariant structure features for reranking showed good effects, however, the features were defined or labeled in a phonetic-context-independent way. In this paper, use of phonetic context to define invariant structure features is examined. The proposed method is tested in two tasks of continuous digits speech recognition and large vocabulary continuous speech recognition (LVCSR). The performances are improved relatively by 4.7% and 1.2%, respectively.","PeriodicalId":248977,"journal":{"name":"2014 IEEE China Summit & International Conference on Signal and Information Processing (ChinaSIP)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116039805","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}