Feature selection is a process to reduce the dimension of a dataset by removing redundant features, and to use the optimal subset of features for machine learning or data mining algorithms. This helps to minimize the time requirement to train a learning algorithm as well as to lessen the storage requirement by ignoring the less-informative features. Feature selection can be considered as a combinatorial optimization problem. In this paper, the authors have presented a new feature selection algorithm called Mayfly-Harmony Search (MA-HS) based on two meta-heuristics namely Mayfly Algorithm and Harmony Search. Mayfly Algorithm has not hitherto been used for feature selection problems to the best of the author’s knowledge. An S-shaped transfer function is incorporated for converting it into a binary version of Mayfly Algorithm. When different candidate solutions obtained from various regions of the search space using Mayfly Algorithm are taken into the harmony memory and processed by Harmony Search, a superior solution can be ensured. This is the primary reason for proposing a hybrid of Mayfly Algorithm and Harmony Search. Thus, combining harmony search with Mayfly Algorithm leads to an increased exploitation of the search space and an overall improvement in the performance of Mayfly-Harmony Search (MA-HS) algorithm. The proposed algorithm has been applied on 18 UCI datasets and compared with 12 other state-of-the-art meta-heuristic FS methods. Experiments have also been performed on three high-dimensional microarray datasets. The results obtained support the superior performance of the algorithm compared to the other methods. The source code of the proposed algorithm can be found using the link as follows: https://github.com/trin07/MA-HS.

Under the present circumstances, when we are still under the threat of different strains of coronavirus, and since the most widely used method for COVID-19 detection, RT-PCR is a tedious and time-consuming manual procedure with poor precision, the application of Artificial Intelligence (AI) and Computer-Aided Diagnosis (CAD) is inevitable. Though, some vaccines have now been authorized worldwide, it will take huge time to reach everyone, especially in developing countries. In this work, we have analyzed Chest X-ray (CXR) images for the detection of the coronavirus. The primary agenda of this proposed research study is to leverage the classification performance of the deep learning models using ensemble learning. Many papers have proposed different ensemble learning techniques in this field, some methods using aggregation functions like Weighted Arithmetic Mean (WAM) among others. However, none of these methods take into consideration the decisions that subsets of the classifiers take. In this paper, we have applied Choquet integral for ensemble and propose a novel method for the evaluation of fuzzy measures using coalition game theory, information theory, and Lambda fuzzy approximation. Three different sets of fuzzy measures are calculated using three different weighting schemes along with information theory and coalition game theory. Using these three sets of fuzzy measures, three Choquet integrals are calculated and their decisions are finally combined. Besides, we have created a database by combining several image repositories developed recently. Impressive results on the newly developed dataset and the challenging COVIDx dataset support the efficacy and robustness of the proposed method. Our experimental results outperform many recently proposed methods.

Non-linear activation functions are integral parts of deep neural architectures. Given the large and complex dataset of a neural network, its computational complexity and approximation capability can differ significantly based on what activation function is used. Parameterizing an activation function with the introduction of learnable parameters generally improves the performance. Herein, a novel activation function called Sinu-sigmoidal Linear Unit (or SinLU) is proposed. SinLU is formulated as $SinLU\left(x\right)=(x+asinbx)·\sigma \left(x\right)$, where $\sigma \left(x\right)$ is the sigmoid function. The proposed function incorporates the sine wave, allowing new functionalities over traditional linear unit activations. Two trainable parameters of this function control the participation of the sinusoidal nature in the function, and help to achieve an easily trainable, and fast converging function. The performance of the proposed SinLU is compared against widely used activation functions, such as ReLU, GELU and SiLU. We showed the robustness of the proposed activation function by conducting experiments in a wide array of domains, using multiple types of neural network-based models on some standard datasets. The use of sine wave with trainable parameters results in a better performance of SinLU than commonly used activation functions.

Time series forecasting is an essential and challenging task, especially for large-scale time-series (LSTS) forecasting, which plays a crucial role in many real-world applications. Due to the instability of time series data and the randomness (noise) of their characteristics, it is difficult for polynomial neural network (PNN) and its modifications to achieve accurate and stable time series prediction. In this study, we propose a novel structure of hierarchical neural networks (HNN) realized by long short-term memory (LSTM), two classes of self-organizing dynamical fuzzy neural network architectures of fuzzy rule-based polynomial neurons (FPN) and polynomial neurons (PN) constructed by variant generation of nodes as well as layers of networks. The proposed HNN combines the deep learning method with the PNN method for the first time and extends it to time series prediction as a modification of PNN. LSTM extracts the temporal dependencies present in each time series and enables the model to learn its representation. FPNs are designed to capture the complex non-linear patterns present in the data space by utilizing Fuzzy C-Means clustering (FCM) and least square error (LSE)-based learning of polynomial functions. The self-organizing hierarchical network architecture generated by the Elitism-based Roulette Wheel Selection (ERWS) strategy ensures that candidate neurons exhibit sufficient fitting ability while enriching the diversity of heterogeneous neurons, addressing the issue of multicollinearity and providing opportunities to select better prediction neurons. In addition, L2-norm regularization is applied to mitigate the overfitting problem. Experiments are conducted on 9 real-world LSTS datasets including three practical applications. The results show that the proposed model exhibits high prediction performance, outperforming many state-of-the-art models.

Facial micro expressions are brief, spontaneous, and crucial emotions deep inside the mind, reflecting the actual thoughts for that moment. Humans can cover their emotions on a large scale, but their actual intentions and emotions can be extracted at a micro-level. Micro expressions are organic when compared with macro expressions, posing a challenge to both humans, as well as machines, to identify. In recent years, detection of facial expressions are widely used in commercial complexes, hotels, restaurants, psychology, security, offices, and education institutes. The aim and motivation of this paper are to provide an end-to-end architecture that accurately detects the actual expressions at the micro-scale features. However, the main research is to provide an analysis of the specific parts that are crucial for detecting the micro expressions from a face. Many states of the art approaches have been trained on the micro facial expressions and compared with our proposed Lossless Attention Residual Network (LARNet) approach. However, the main research on this is to provide analysis on the specific parts that are crucial for detecting the micro expressions from a face. Many CNN-based approaches extracts the features at local level which digs much deeper into the face pixels. However, the spatial and temporal information extracted from the face is encoded in LARNet for a feature fusion extraction on specific crucial locations, such as nose, cheeks, mouth, and eyes regions. LARNet outperforms the state-of-the-art methods with a slight margin by accurately detecting facial micro expressions in real-time. Lastly, the proposed LARNet becomes accurate and better by training with more annotated data.