A Neural Architecture for Multi-label Text Classification

On the methodology of robotic design from the traditional view of communication which is assumed as a symbol process, robots are forced to confront the symbol grounding problem. However, if communication is assumed as the analog dynamics and robots are driven by it, robots can avoid the problem and be situated in the environment and to other agents. In this paper we introduce a new communication system constructed from the view of dynamical systems to achieve the situatedness. This system is that there is a robot in a virtual environment and the control of the robot is shared by human operation using a joystick and a robot controller. As the controller, we adopt multiple recurrent neural networks (MRNN) which are able to cope with complex environments and broad communication that single recurrent net cannot cope with. We conduct two experiments in order to evaluate the effectiveness of MRNN to a low level communication task such as nonverbal interaction. First, we examine the effect of the number of RNNs contained in MRNN. Second, we examine the effect of the context dependency of MRNN. These experiments show the capability of MRNN as a new-type controller of communication robot.

Time-delay recurrent neural network for temporal correlations and prediction

In multi-label classification a datapoint can be labelled with more than one class at the same time. A common but trivial approach to multi-label classification is to train individual binary classifiers per label, but the performance can be improved by considering associations between the labels, and algorithms like classifier chains and RAKEL do this effectively. Like most machine learning algorithms, however, these approaches require accurate hyperparameter tuning, a computationally expensive optimisation problem. Tuning is important to train a good multi-label classifier model. There is a scarcity in the literature of effective multi-label classification approaches that do not require extensive hyperparameter tuning. This paper addresses this scarcity by proposing CascadeML, a multi-label classification approach based on cascade neural network that takes label associations into account and requires minimal hyperparameter tuning. The performance of the CasecadeML approach is evaluated using 10 multi-label datasets and compared with other leading multi-label classification algorithms. Results show that CascadeML performs comparatively with the leading approaches but without a need for hyperparameter tuning.

Based on principal component analysis, a multiple neural network was proposed. The principal component analysis was firstly used to reorganize the input variables and eliminate the correlativity. Then the reorganized variables were divided into 2 groups according to the original information and 2 corresponding neural networks were established. A radial basis function network was used to depict the relationship between the output variables and the first group input variables which contain main original information. An other single-layer neural network model was used to compensate the error between the output of radial basis function network and the actual output variables. At last, The multiple network was used as soft sensor for the ratio of soda to aluminate in the process of high-pressure digestion of alumina. Simulation of industry application data shows that the prediction error of the model is less than 3%, and the model has good generalization ability.

Distributed [formula omitted]-winners-take-all via multiple neural networks with inertia

Improved on-line process fault diagnosis through information fusion in multiple neural networks

PurposeImage classification is becoming a supporting technology in several image-processing tasks. Due to rich semantic information contained in the images, it is very popular for an image to have several labels or tags. This paper aims to develop a novel multi-label classification approach with superior performance.Design/methodology/approachMany multi-label classification problems share two main characteristics: label correlations and label imbalance. However, most of current methods are devoted to either model label relationship or to only deal with unbalanced problem with traditional single-label methods. In this paper, multi-label classification problem is regarded as an unbalanced multi-task learning problem. Multi-task least-squares support vector machine (MTLS-SVM) is generalized for this problem, renamed as multi-label LS-SVM (ML2S-SVM).FindingsExperimental results on the emotions, scene, yeast and bibtex data sets indicate that the ML2S-SVM is competitive with respect to the state-of-the-art methods in terms of Hamming loss and instance-based F1 score. The values of resulting parameters largely influence the performance of ML2S-SVM, so it is necessary for users to identify proper parameters in advance.Originality/valueOn the basis of MTLS-SVM, a novel multi-label classification approach, ML2S-SVM, is put forward. This method can overcome the unbalanced problem but also explicitly models arbitrary order correlations among labels by allowing multiple labels to share a subspace. In addition, the multi-label classification approach has a wider range of applications. That is to say, it is not limited to the field of image classification.

Contributions of MIR to soundscape ecology. Part 3: Tagging and classifying audio features using a multi-labeling k-nearest neighbor approach

Artificial neural network is widely used in applications including image classification. However, due to the limitation and drawbacks of a single neural network, the accuracy of the image recognition can be further improved by using multiple neural networks that have different focuses in the method. In this paper, a conventional neural network based on the shape mask of the object is first used to classify fashion images. Then a convolutional neural network (CNN) is applied to identify the same set of images. In order to take advantage of different neural network approaches, a weight method is proposed in this paper. By considering the recognition accuracy of each class for each approach, a fused decision is made. Experimental results show that the fusion of multiple neural networks provides better overall accuracy than the individual approach, and achieves a lower number of false classifications in five out of ten classes.

<p>Biological processes are highly nonlinear in nature and difficult to represent accurately by simple mathematical models. However, this problem can be solved by using neural network. Neural network is a prominent modeling tool especially when it comes to intricate process such as biological process. In this paper, a multiple single hidden layer with ten hidden neurons Feedforward Artificial Neural Network (FANN) was used to model the complex and dynamic relationships between the input (dilution rate, D) and outputs (conversion, y and dimensionless temperature value, θ) for the reactive biological process. Levenberg-Marquardt Backpropagation training method was used. The multiple neural networks predicted outputs were then combined through three different methods which area simple averaging, Principal Component Regression (PCR) and Independent Component Regression (ICR). Multiple neural networks which were created by the bootstrap approach help improved single neural network performance as well as the model robustness for nonlinear process modeling. Comparison was made between the three methods. The result showed that ICR is slightly superior between the three methods especially in noise level 1,2 and 3, however ICR slightly suffer in noise level 4 and 5. This is due to the independent component regression used the latent factors and non-Gaussian distribution of y and θ values for the combination.</p><p>Chemical Engineering Research Bulletin 19(2017) 12-19</p>

Problem transformation and algorithm adaptation are the two main approaches in machine learning to solve multilabel classification problem. The purpose of this paper is to investigate both approaches in multilabel classification for Indonesian news articles. Since this classification deals with a large number of features, we also employ some feature selection methods to reduce feature dimension. There are four factors as the focuses of this paper, i.e., feature weighting method, feature selection method, multilabel classification approach, and single-label classification algorithm. These factors will be combined to determine the best combination. The experiments show that the best performer for multilabel classification of Indonesian news articles is the combination of TF-IDF feature weighting method, Symmetrical Uncertainty feature selection method, Calibrated Label Ranking — which belongs to problem transformation approach —, and SVM algorithm. This best combination achieves F-measure of 85.13% in 10-fold cross-validation, but the F-measure decreases to 76.73% in testing because of OOV.

An innovative ensemble model based on multiple neural networks and a novel heuristic optimization algorithm for COVID-19 forecasting

This paper presents an off-line signature verification system. The verification is based on signature shape descriptors such as the skeleton, upper and lower envelopes, and the high pressure region of the signatures. Multiple multilayer perceptron neural network modules cooperating in taking a verification decision via a fuzzy integral voter are used. This method is capable of verifying simple and skilled forgeries with a good performance.

A combinatorial optimization approach for multi-label associative classification

Experiment with intensity level for the LC30 ceramsite concrete as the research object, changing the content of cement, GRT fiber, rubber powder by the orthogonal test to configure GRT fiber—rubberized haydite concrete samples, maintenance samples 7d and 28d in standard conditions and respectively testing their standard compressive strength. Through the analysis of the test data, using multiple regression analysis established the GRT fiber—rubberized haydite concrete 7d and 28d standard compressive strength regression formulas.By means of BP neural network theory combine MATLAB programme established GRT fiber—rubberized haydite concrete 7d and 28d standard compressive strength neural network model.Finally using 3 groups new test data to compare the value of multiple regression equations and BP neural network’s predicted value.The results indicate that the multiple regression equations and BP neural network model are availabled.