Semi-supervised learning using multiple clusterings with limited labeled data

Semi-Supervised k-Star (SSS): A Machine Learning Method with a Novel Holo-Training Approach

As one of the entropy-based methods, the k-Star algorithm benefits from information theory in computing the distances between data instances during the classification task. k-Star is a machine learning method with a high classification performance and strong generalization ability. Nevertheless, as a standard supervised learning method, it performs learning only from labeled data. This paper proposes an improved method, called Semi-Supervised k-Star (SSS), which makes efficient predictions by considering unlabeled data in addition to labeled data. Moreover, it introduces a novel semi-supervised learning approach, called holo-training, against self-training. It has the advantage of enabling a powerful and robust model of data by combining multiple classifiers and using an entropy measure. The results of extensive experimental studies showed that the proposed holo-training approach outperformed the self-training approach on 13 out of the 18 datasets. Furthermore, the proposed SSS method achieved higher accuracy (95.25%) than the state-of-the-art semi-supervised methods (90.01%) on average. The significance of the experimental results was validated by using both the Binomial Sign test and the Friedman test.

DCA for Sparse Quadratic Kernel-Free Least Squares Semi-Supervised Support Vector Machine

With the development of science and technology, more and more data have been produced. For many of these datasets, only some of the data have labels. In order to make full use of the information in these data, it is necessary to classify them. In this paper, we propose a strong sparse quadratic kernel-free least squares semi-supervised support vector machine (SSQLSS3VM), in which we add a ℓ0norm regularization term to make it sparse. An NP-hard problem arises since the proposed model contains the ℓ0 norm and another nonconvex term. One important method for solving the nonconvex problem is the DC (difference of convex function) programming. Therefore, we first approximate the ℓ0 norm by a polyhedral DC function. Moreover, due to the existence of the nonsmooth terms, we use the sGS-ADMM to solve the subproblem. Finally, empirical numerical experiments show the efficiency of the proposed algorithm.

Densely connected attention network for diagnosing COVID-19 based on chest CT

BACKGROUND

To fully enhance the feature extraction capabilities of deep learning models, so as to accurately diagnose coronavirus disease 2019 (COVID-19) based on chest CT images, a densely connected attention network (DenseANet) was constructed by utilizing the self-attention mechanism in deep learning.

METHODS

During the construction of the DenseANet, we not only densely connected attention features within and between the feature extraction blocks with the same scale, but also densely connected attention features with different scales at the end of the deep model, thereby further enhancing the high-order features. In this way, as the depth of the deep model increases, the spatial attention features generated by different layers can be densely connected and gradually transferred to deeper layers. The DenseANet takes CT images of the lung fields segmented by an improved U-Net as inputs and outputs the probability of the patients suffering from COVID-19.

RESULTS

Compared with exiting attention networks, DenseANet can maximize the utilization of self-attention features at different depths in the model. A five-fold cross-validation experiment was performed on a dataset containing 2993 CT scans of 2121 patients, and experiments showed that the DenseANet can effectively locate the lung lesions of patients infected with SARS-CoV-2, and distinguish COVID-19, common pneumonia and normal controls with an average of 96.06% Acc and 0.989 AUC.

CONCLUSIONS

The DenseANet we proposed can generate strong attention features and achieve the best diagnosis results. In addition, the proposed method of densely connecting attention features can be easily extended to other advanced deep learning methods to improve their performance in related tasks.

Predicting Alcohol Concentration during Beer Fermentation Using Ultrasonic Measurements and Machine Learning

Beer fermentation is typically monitored by periodic sampling and off-line analysis. In-line sensors would remove the need for time-consuming manual operation and provide real-time evaluation of the fermenting media. This work uses a low-cost ultrasonic sensor combined with machine learning to predict the alcohol concentration during beer fermentation. The highest accuracy model (R2 = 0.952, mean absolute error (MAE) = 0.265, mean squared error (MSE) = 0.136) used a transmission-based ultrasonic sensing technique along with the measured temperature. However, the second most accurate model (R2 = 0.948, MAE = 0.283, MSE = 0.146) used a reflection-based technique without the temperature. Both the reflection-based technique and the omission of the temperature data are novel to this research and demonstrate the potential for a non-invasive sensor to monitor beer fermentation.

Exploiting the interplay between cross-sectional and longitudinal data in Class III malocclusion patients

The aim of the study was to investigate how to improve the forecasting of craniofacial unbalance risk during growth among patients affected by Class III malocclusion. To this purpose we used computational methodologies such as Transductive Learning (TL), Boosting (B), and Feature Engineering (FE) instead of the traditional statistical analysis based on Classification trees and logistic models. Such techniques have been applied to cephalometric data from 728 cross-sectional untreated Class III subjects (6-14 years of age) and from 91 untreated Class III subjects followed longitudinally during the growth process. A cephalometric analysis comprising 11 variables has also been performed. The subjects followed longitudinally were divided into two subgroups: favourable and unfavourable growth, in comparison with normal craniofacial growth. With respect to traditional statistical predictive analytics, TL increased the accuracy in identifying subjects at risk of unfavourable growth. TL algorithm was useful in diffusion of information from longitudinal to cross-sectional subjects. The accuracy in identifying high-risk subjects to growth worsening increased from 63% to 78%. Finally, a further increase in identification accuracy, up to 83%, was produced by FE. A ranking of important variables in identifying subjects at risk of growth worsening, therefore, has been obtained.