China
Africa
Explore chapters and articles related to this topic
Implementation of Data-Driven Approaches for Condition Assessment of Structures and Analyzing Complex Data
Published in M.Z. Naser, Leveraging Artificial Intelligence in Engineering, Management, and Safety of Infrastructure, 2023
Vafa Soltangharaei, Li Ai, Paul Ziehl
In physics-based methods, the availability of a physical model and loading conditions are essential. Each model should be validated and its parameters determined before application, which is not always feasible and straightforward (An et al., 2015). Computational intelligence, machine learning, and regression are three examples of data-driven methods (Solomatine et al., 2009). Pattern recognition can be categorized under the machine learning field, which focuses on discovering the regularities inside a data set by employing computer algorithms. There are two main categories associated with this method: unsupervised and supervised pattern recognition. Unsupervised pattern recognition is utilized to find and cluster unlabeled data when no predefined pattern is available. On the other hand, supervised pattern recognition is employed when a large labeled dataset is accessible for training. Subsequently, the trained algorithm can be engaged to classify new data (Bishop, 2006).
Data Analysis
Published in Paul L. Goethals, Natalie M. Scala, Daniel T. Bennett, Mathematics in Cyber Research, 2022
Raymond R. Hill, Darryl K. Ahner
Any book chapter is necessarily limited. This chapter is no exception. This chapter examined a representative sample of data analytical methods aligned with the classification provided in Table 8.1. The discussion was not complete. For instance, there are other classification methods, much more than the three discussed here. For example, the Naïve Bayes method or the use of neural networks for classification. There are also a wide variety of pattern recognition algorithms, none of which we discuss here. Those pattern recognition algorithms include neural networks, wavelet models, and a variety of deep-learning network models. For instance, Butt (2018) used neural networks for cyber attack detection. Bishop (2016) is a well cited source text on this topic. In general, there are a plethora of excellent books, papers, and available tutorials on each of the subjects. For instance the encyclopedia by Sammut and Webb (2011) on the variety of machine learning topics, the introductory book by Manley and Alberto (2017) and the more advanced text by James, Witten, Hastie, and Tibshirani (2013).
Application of machine learning and deep learning in cybersecurity
Published in Sabyasachi Pramanik, Anand Sharma, Surbhi Bhatia, Dac-Nhuong Le, An Interdisciplinary Approach to Modern Network Security, 2022
Dushyant Kaushik, Muskan Garg, Annu, Ankur Gupta, Sabyasachi Pramanik
In these three methods, ML/DL are comparable: supervised, unmonitored, and semi-supervised. Each instance consists of an input sample and a mark during supervised learning. The supervised learning algorithm analyses the data from the training and maps new instances using the results of the study. Unsupervised learning from unlabeled data deduces the definition of secret structures. Since the dataset is unclassified, the precision of the performance of the algorithm cannot be checked, and it is possible to summarize and describe only the key features of the data. Semi-supervised learning is a way to blend supervised and unsupervised learning. Since the dataset is unlabeled, the precision of the performance of the algorithm cannot be checked, so it is only possible to summarize and describe those features. A means of mixing supervised learning and unsupervised learning is semi-supervised learning. Unlabeled data is used by semi-supervised learning when using labeled data for pattern recognition. Using semi-supervised education will decrease efforts to mark, thus attaining high precision.
Current and future role of data fusion and machine learning in infrastructure health monitoring
Published in Structure and Infrastructure Engineering, 2023
Hao Wang, Giorgio Barone, Alister Smith
Pattern recognition refers to techniques used to identify patterns and regularities in data. Due to the increased availability of data and abundance of computational power, ML techniques have been widely used for pattern recognition in a range of disciplines, including statistics, engineering, signal processing and medical science, over the past few decades. In regards to the assessment of the infrastructure health condition, ML techniques are today one of the most widely used computational approaches due to their ability in handling data with large numbers of redundancies and uncertainties (Chen & Gu, 2019). As infrastructure health monitoring systems aim to recognise damages in infrastructures, which can be usually identified as unusual patterns, both supervised and unsupervised learning are appropriate. The following subsections focus on a variety of techniques of supervised learning, unsupervised learning and decision-level fusion, and their applications in infrastructure health monitoring.
Safe transductive support vector machine
Published in Connection Science, 2022
Haiyan Chen, Ying Yu, Yizhen Jia, Linghui Zhang
In the real world, there are a large number of unlabelled samples which are difficult to obtain labels. Therefore, they cannot be directly used for machine learning. For example, in the field of sports bioanalysis (Honglian et al., 2020; Huifeng et al., 2020), there are a large number of samples collected from the real world without any labels. Before these samples are used for learning, we have to ask sports biologists to manually label these samples, which is a costly task. The same problem also exists in other fields, such as cross-domain sentiment classification (Cao et al., 2021), human cognition (Thibodeau et al., 2020) and social analysis (Dai & Wang, 2021; Qiu et al., 2021). Therefore, semi-supervised learning (Chapelle et al., 2009; Zhu & Goldberg, 2009) which can use a large number of unlabelled samples and a small number of labelled samples to train a better model has become a key approach in pattern recognition and machine learning.
Client profile prediction using convolutional neural networks for efficient recommendation systems in the context of smart factories
Published in Enterprise Information Systems, 2022
Nadia Nedjah, Victor Ribeiro Azevedo, Luiza De Macedo Mourelle
Deep learning is machine learning category, wherein complex neural network models are exploited to recognise patterns in very large datasets, commonly known as big data. This model is used in many applications, especially those dedicated to pattern recognition, such as nodules detection (Pinheiro, Nedjah, and Mourelle 2020), intrusion detection (Patil and Patil 2019). Pattern recognition in images and videos can be considered one of the most widely used applications of deep learning. This work contributes to the scientific community, in the field of computational vision and image classification, with exploratory and experimental analysis of the performance of Convolutional Neural Networks (CNNs). Different training algorithms are applied to a group of images in the categories: animals, electronics, games, vehicles and clothes. To improve the performance of the deep learning system, experiments are performed varying the structure of deep learning training algorithms and the initialisation of hyper-parameters, such as the number of epochs and the rate of learning. In this work, we apply and investigate the performance of the following training algorithms: SGD, AdaGrad, Adamax, RMSprop and Adam.