China 
Africa 
Explore chapters and articles related to this topic
Machine Learning/Deep Learning for Natural Disasters
Published in Sam Goundar, Archana Purwar, Ajmer Singh, Applications of Artificial Intelligence, Big Data and Internet of Things in Sustainable Development, 2023
Tripti Sharma, Anjali Singhal, Kumud Kundu, Nidhi Agarwal
An important algorithm, Akaike information concept (AIC), is used widely among the research community for earthquake detection. One of the major drawbacks of this algorithm is that it is unable to estimate the minimal globally incorrect values when the backdrop of the noisy disturbance increases (Leonard, M. et al. 1999). In order to overcome these disadvantages a group of researchers suggested automatic detection of earthquakes using various machine learning methods like K-means algorithm (Chai, X. et.al 2020, Chen, Y. 2018), templates match method (Beroza, G.C. 2019), support vector machine (Chen, Y. 2020; Aad, O.M. et al. 2020a), fuzzification algorithms (Chen, Y. 2020), wavlytic transformational changes (Hafez, A.G. et al. 2013), and various threshold controlling algorithms. As discussed earlier in Jozinovic, D. (2019) and Harirchian, E.ALAMR et al. (2020), various deep learning algorithms are gaining more importance for early detection and prediction of earthquakes. Deep learning-based algorithms are used like CNN, RNN, and auto-encoder (Zhu, W. and Saad, O.M. et al. 2018; Beroza, G.C. et al., 2019, Evries, P.M. 2018). The most important aspect of using CNN is that it is capable of extracting the most significant attributes from the input data. But it suffers from various shortcomings also. The Mousavi, S. M. et al. (2019) various layers of CNN architecture do not completely contribute to enhancing the dimensional value of the network. There is one layer present in the CNN architectural framework that is called the pool layer, which reduces the dimensional value of the whole structure. The prediction model (due to pool layer presence) does not have the scope to lose important data. This problem reduces the predictive performance of the CNN algorithm (Sabour, G.E. et al. 2017; Jia et al. 2020). Capsule neural network, which was suggested for the first time by Sabour et al. (2017), was helpful in removing the drawbacks of the CNN algorithm. Capsule neural network is the improved technology for the deep learners’ architecture, which provides the ability to the framework to adapt and enhance learning even with the absence of a pool layer. In such a case it proceeds without the loss of any data and provides training to the model with even a lesser number of samples of data to finally achieve generalized enhanced capability. It is able to achieve remarkable improvements as it uses a vectored output for the simple network framework. This technology is being used efficiently by many researchers inclined toward the earthquake seismic data analysis from various areas like South California (Ross, Z.E. et al. 2018), Asia and Europe. The end results are obtained with better predictive performance accuracy than the various other existing algorithms used in previous studies (Ross, Z.E. et al. 2018).
An optimal detection of fake news from Twitter data using dual-stage deep capsule autoencoder
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2023
Parimala Kanaga Devan K, Anandha Mala G S
Detection of fake news from Twitter data is a challenging problem faced by several existing techniques. Thus, to attain superior classification results, the proposed study prefers a hybrid deep learning model for fake news detection and classification. Recently, the capsule neural network has attained greater attention for text classification. This capsule network can acquire more essential features and spatial information from the text. To classify the input tweet text as real or fake, fetching essential features is very important. The capsule network has been designed to overcome the drawbacks of the traditional CNN network. Various existing studies used capsule networks and gained better outcomes. Like the capsule network, the autoencoder is also an effective technique for text classification. The classification process mainly reduces outcomes for larger dimensionality data because of redundant information. However, this autoencoder performs an encoding process in which the minimal dimensional representation is learned for larger dimensional data. So, this autoencoder method reduces data’s dimensionality and captures the needed information from the given text. Thus, by considering the benefits of these deep learning models, the proposed study hybridises them for fake news detection and classification. The selected features from the HOA approach are the input of the proposed classification stage. The proposed DSDC-AE-based Twitter data classification model involves two stages: encoding and decoding. In the proposed Twitter data classification, the capsule encoder is utilised for the encoding process, and the autoencoder is used for the decoding process. The encoding process gathers the input from the HOA approach and converts it into a compressed form. The compressed form of the input features is fed to the autoencoding-based decoder in which the reconstruction of the real input is obtained. The evaluation from input state to hidden state and from hidden state to output state is said as the process of encoding and decoding.
Analyzing the performances of squash functions in capsnets on complex images
Published in Cogent Engineering, 2023
Benjamin A. Weyori, Yaw Afriyie, Alex A. Opoku
Hinton’s 2017 paper (Sara et al., 2017) presents the capsule vectors as convolutional architectures, based on the concept of a capsule neural network. An alternative to traditional down-sampling methods such as max pooling is proposed that selectively links units within a capsule together. In 2018, Hinton published a follow-up article (Hinton et al., 2018) that extended capsules to matrix form and further developed the routing scheme; however, our study will primarily focus on the architecture discussed in the baseline study (Sara et al., 2017), and we will perform experiments using the dynamic routing algorithm(see Algorithm 1) in parallel with those in (Sara et al., 2017). Several other modifications to the original architecture have also been proposed, such as in (Edgar et al., 2017; Yaw et al., 2022a), where the number of layers and capsule size is increased as well as changes to the activation function is made. Although the dynamic routing procedure recently proposed by (Sara et al., 2017) is effective, there is no standard formalization of the heuristic. According to (Wang & Liu, 2018), the routing strategy proposed by (Sara et al., 2017) can be partially expressed as an optimization problem that minimizes a clustering-like loss and a KL regularization term between the current coupling distribution and its last state. In addition, the authors introduce another simple routing method that exhibits a number of interesting features. As described in (Rawlinson et al., 2018), capsules without masking may be more generalizable than those with masking. According to (Martins et al., 2019), multi-lane capsule networks (MLCNs) are a resource-efficient way to organize capsule networks (CapsNets) for parallel processing and high accuracy at low costs. With CapsNet, MLCNs consist of several (distinct) parallel lanes, each contributing to a dimension of the result. In both Fashion-MNIST and CIFAR-10 datasets, their results indicate similar accuracy with reduced parameter costs. In addition, when using a proposed novel configuration for the lanes, the MLCN outperforms the original CapsNet. Furthermore, MLCN has faster training and inference times than CapsNet in the same accelerator, over twofold faster. By combining pairwise inputs with the capsule architecture, the authors in (Neill, 2018) construct a Siamese capsule network. Siamese Capsule Networks outperform strong baselines on two pairwise learning datasets, exhibiting the greatest performance in the few-shot learning setting where pairwise images contain unseen subjects.