查看更多>>摘要：Adversarial Multi-Task Learning (AMTL) has demonstrated its promising capability of information capturing and representation learning, however, is hardly explored in speech enhancement. In this paper, we propose a novel adversarial multi-task learning with inverse mapping method for speech enhancement. Our method focuses on enhancing the generator's capability of speech information capturing and representation learning. To implement this method, two extra networks (namely P and Q) are developed to establish the inverse mapping from the generated distribution to the input data domains. Correspondingly, two new loss functions (i.e., latent loss and equilibrium loss) are proposed for the inverse mapping learning and the enhancement model training with the original adversarial loss. Our method obtains the state-of-the-art performance in terms of speech quality (PESQ=2.93, CVOL=3.55). For speech intelligibility, our method can also obtain competitive performance (STOI=0.947). The experimental results demonstrate that our method can effectively improve speech representation learning and speech enhancement performance. (c) 2022 Elsevier B.V. All rights reserved.
Varshney, Ayush K.Muhuri, Pranab K.Lohani, Q. M. Danish
15页
查看更多>>摘要：Hierarchical clustering techniques help in building a tree-like structure called dendrogram from the data points which can be used to find the closest related data objects. This paper presents a novel hierarchical clustering technique which considers intuitionistic fuzzy sets to deal with the uncertainty present in the data. Instead of using traditional hamming distance or Euclidean distance measure to find the distance between the data points, it employs the probabilistic Euclidean distance measure to propose a novel clustering approach which we term as 'Probabilistic Intuitionistic Fuzzy Hierarchical Clustering (PIFHC) Algorithm'. The proposed PIFHC algorithm considers probabilistic weights from the data to measure the distances between the data points. Clustering results over UCI datasets show that our proposed PIFHC algorithm gives better cluster accuracies than its existing counterparts. PIFHC efficiently provides improvements of 1%-3.5% in the clustering accuracy compared to other fuzzy hierarchical clustering algorithms for most of the datasets. We further provide experimental results with the real-world car dataset and the Listeria monocytogenes dataset for mouse susceptibility to demonstrate the practical efficacy of the proposed algorithm. For Listeria datasets as well, proposed PIFHC records 1.7% improvement against the state-of-the-art methods The dendrograms formed by the proposed PIFHC algorithm exhibits high cophenetic correlation coefficient with an improvement of 0.75% over others. We provide various AGNES methods to update the distance between merged clusters in the proposed PIFHC algorithm. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：In wastewater treatment processes (WWTPs), the effluent ammonia nitrogen (NH4-N) concentration is a significant index to measure effluent quality. Recently, the soft computing methods have been widely used to measure effluent NH4-N concentration. However, the performance of soft computing method is closely related with its input variables, which is difficult to choose. As an alternative, the time series prediction method PSR-PRWNN is proposed, which combines phase space reconstruction (PSR) technique and pipelined recurrent wavelet neural network (PRWNN). Different from soft computing methods, the time series prediction method is a method which predicts the effluent NH4-N concentration by using its history data rather than other variables data. Firstly, the chaotic characteristics of effluent NH4-N time series is proved by using the correlation dimension method. Based on chaotic characteristics, the phase space of effluent NH4-N concentration is reconstructed by PSR technique. Then, the relationship model between inputs and output of the reconstructed phase space is established by PRWNN. Thirdly, the parameters of PRWNN are trained by an online gradient algorithm with adaptive learning rates. Finally, the experimental results indicate that the PSR-PRWNN can obtain better training results and prediction accuracy than other algorithms. (c) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：SMOTE is a well-known oversampling method for learning on imbalanced datasets. However, it has the risk of introducing noisy instances and overfitting problems. In order to improve its performance, this paper proposes an oversampling method called SMOTE-COF, which is an improvement of SMOTE based on center offset factor. The SMOTE-COF method first removes noisy samples, then computes center offset factor to select sparsely distributed minority class samples. Furthermore, these samples are used to generate new minority class samples with other minority class instances distributed in the same sub-cluster by SMOTE. Comparative experiments on one simulated dataset and fourteen UCI datasets provide evidence that the SMOTE-COF can effectively reduce noisy samples, generate better minority classes, and improve classification performance for imbalanced datasets. (c) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：Object recognition under occlusion is a key issue in computer vision. Since one can recognize an occluded object solely based on the shape, one ultimate goal of artificial intelligence is to find an automatic method that could recognize the object solely based on its shape with equal recognition accuracy. In this paper, slope difference distribution (SDD) is used to extract the shape features of the object as its sparse representation. One or several scale-invariant shape models are defined with the general SDD features for each shape class. The object is recognized based on the minimum distances between its detected SDD features and the SDD features of all the shape models. To increase the generality, we propose a two-dimensional SDD feature extraction method that computes the SDD features directly from the two-dimensional contours. Experimental results showed that the proposed object recognition method could recognize the object under significant occlusion robustly. It achieved 100% recognition and retrieval accuracy on three public datasets, Kimia99, Kimia216 and MPEG-7. For the fine-grained object classification, the proposed method achieved 90.6% accuracy on CUB-200-2011, which is also better than existing methods. (c) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：Conventional application of deep neural networks (DNNs) to multi-building and multi-floor indoor localization is based on pure regression of three-dimensional location coordinates (e.g., longitude, latitude and altitude (i.e., floor height)), classification of location labels (e.g., building, floor and room information), or hybrid classification/regression of labels and coordinates (e.g., building and floor information and two-dimensional location coordinates), which, however, does not take into account an innate hierarchical auxiliary information (e.g., building->floor->location) of indoor localization data. Such conventional application of DNNs faces scalability issues in case of large-scale indoor localization where the numbers of buildings and floors are large. Inserting classification tasks as auxiliary networks into a regression neural network, we propose a new framework called a hierarchical auxiliary deep neural network (HADNN), which not only address the scalability issues with an increasing number of classes but also could further reduce the hierarchical information error. In HADNN, hierarchical auxiliary information of given data are provided and used during the training phase. As there are two possible hierarchical information cases in indoor localization data: (1) given only floors and (2) given both buildings and floors, we propose two architectures: one utilizing only floor information and the other taking both building and floor information. At test phase, HADNN predicts building, floor and location coordinate at the same time. Experimental results show that the architecture of HADNN achieves better performance of a coordinate regression task and require a smaller number of parameters than the pure two-dimensional location coordinates regression model. In addition, HADNN does not require the training data and coarse classes (e.g., building and floor information) at test phase while previous methods still require the training data to obtain location coordinates. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：Concerning the problems of weak scalability of traditional collaborative filtering recommender systems, a scalable recommender system based on bi-clustering and moth flame optimization algorithm is proposed. First of all, the users-items scoring matrix is filtered and cleaned in order to reduce the computational overhead, afterwards the bi-clustering data structures are constructed for the processed matrix, and the algorithm searches for bi-cluster containing the target user. Then, the results of biclustering are set as the initial population, and the moth flame optimization algorithm is applied to deeply optimize the similar users. Finally, the unrated items are predicted for the target user, and the recommendation list is generated for the target user. Validation experiments are carried on different scales of datasets; the results show that the proposed system achieves a good scalability, and also good recommendation performance. (c) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：From collected experimental data, a rapid and precise classification model for impact damage modes (IMDs) can be developed using machine learning (ML) techniques to evaluate impact resistant capabilities of reinforced concrete (RC) building walls. However, experimental data is often small and imbalanced, resulting in significant degradation and instability in classification performance. In this study, an imbalanced 4-classes dataset consisted of 240 missile impact tests is employed, with the most minor class containing only 10 samples. The paper aims to develop an automated classification model for IDMs, using a clustering-based within-class stratified splitting technique, named WICS, combining with a well-known oversampling technique, namely SMOTE-NC, that considers not only the between class imbalance but also the within-class distribution to stabilize the classification performance. Four classifiers and five data splitting techniques are developed and implemented to address classification performance. We found that the support vector machine (SVM) classifier using WICS and SMOTE NC achieves the best micro F1 score (0.821), Cohen's kappa score (0.700), and AUC value (0.949) with highly stable performance. Friedman and Holm's post-hoc statistical tests also confirm the outperformance of WICS+SMOTE-NC over other techniques. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：In this study, binary versions of the Artificial Algae Algorithm (AAA) are presented and employed to determine the ideal attribute subset for classification processes. AAA is a recently proposed algorithm inspired by microalgae's living behavior, which has not been consistently implemented to determine ideal attribute subset (feature selection) processes yet. AAA can effectively look into the feature space for ideal attributes combination minimizing a designed objective function. The proposed binary versions of AAA are employed to determine the ideal attribute combination that maximizes classification success while minimizing the count of attributes. The original AAA is utilized in these versions while its continuous spaces are restricted in a threshold using an appropriate threshold function after flattening them. In order to demonstrate the performance of the presented binary artificial algae algorithm model, an experimental study was conducted with the latest seven highperformance optimization algorithms. Several evaluation metrics are used to accurately evaluate and analyze the performance of these algorithms over twenty-five datasets with different difficulty levels from the UCI Machine Learning Repository. The experimental results and statistical tests verify the performance of the presented algorithms in increasing the classification accuracy compared to other state-of-the-art binary algorithms, which confirms the capability of the AAA algorithm in exploring the attribute space and deciding the most valuable features for classification problems. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：In dynamic environments, such as ground-based optic observations for space targets, anisotropic turbulence and random noise tend to produce different discriminate image degradations at every moment. These degradations severely reduce the quality of the target images and make restoration of these images very difficult. However, an inconsistent degradation implies that there is complementary information in such images. In this paper, a ranking network (Ranknet) is first proposed to ensure that input sequences have a consistent distribution upon degradation and squeeze the spatial distribution of the sample set. Then, an extraction-refinement neural network (ERnet) is proposed to extract complementary features and blindly reconstruct a clean image of the observation target. In ERnet, an extraction subnetwork (EN) uses 3D convolutions to extract discriminate features from multiframe input sequences, and a refinement subnetwork (RN) based on 2D convolution restores clean images by refining the effective features. In addition, a spatial-temporal attention module (STAM) is devoted to enhancing features through utilization of the high-quality features. Experimental results on the restorations of space target images and motion blur images confirm the superior performance of ERnet as compared with other state-of-the-art methods. (c) 2022 Elsevier B.V. All rights reserved.
NSTL
Elsevier