(70b) Deep Learning Applications in Advanced Analytic Approaches of Big Data

Recent versions of industries known as Industry 4.0 and Industry 5.0 are set up on wide ranges of the cutting-edge technologies such as Internet of Things (IoT). IoT is considered as a dynamic global network infrastructure with self-configuring capabilities based on standard and interoperable communication protocols where physical and virtual ‘Things’ have identities, physical attributes, and virtual personalities and use intelligent interfaces [1] [2]. IoT is among those technologies that facilitates data flow lead to making massive volume of data, known as Big Data. Several definitions and explanations are suggested to address Big Data, but it mainly refers to massive datasets that usual hardware and software cannot collect, process, and analyze it [3] [4].

Performing advanced analytic tasks would become more complicated and demanding when they have to cope with inherent Big Data characteristics. Analytic approaches that deal with Big Data aim to identify and predict current and future patterns and trends to describe process situations. Most conventional machine learning algorithms cannot detect complicated hidden patterns and trends in Big Data. In contrast, Deep learning (DL) algorithms have presented superb performance in satisfying this requirement through automating the learning procedure and providing complex, robust models. DL algorithms facilitate being confronted with intrinsically troublemaking aspects of Big Data, which derives from DL aptitudes for analyzing large volumes of data available with a high rate from multiple sources. DL algorithms can learn from unsupervised data, in addition to supervised data, which is a key feature for any algorithm to be applied in advanced analytics approaches.

DL was presented in 2006, however, the roots of the concept date back to the 1940s [5]. DL models are deeper variants of artificial neural networks (ANNs) with multiple layers, whether linear or non-linear. Each layer is connected to its lower and upper layers through different weights. Through attempting to imitate the hierarchical learning structure of human brain, DL algorithms automatically extract patterns, representations, and trends from data. DL, in fact, is a deep architecture of layers that apply nonlinear transformations to input to provide representations of it in output [6]. Two appropriate DL architectures for advanced analytical approaches of Big Data are Deep Belief Network (DBN) and Convolutional Neural Networks (CNN). DBN constructs a model using unsupervised and supervised techniques, consisting of an input layer, hidden layers, and an output layer, while CNN is a supervised technique that has multiple hierarchical layers [7] [8].

Enhanced DL architectures assist advanced analytic approaches to discover meaningful Big Data patterns for making decisions, predictions, and conclusions. As an instance, it is proven that the parameters of large-scale DL models, especially with fully connected layers, are of high redundancy. Therefore, some methods, such as Hashing Trick, have been presented to improve the training efficiency by significantly compressing the parameters without a noteworthy accuracy drop. In addition, some software such as DistBelief is developed to train large-scale DL models in multiple machines in parallel. Through partitioning a large deep model, it significantly improves the training efficiency [5].

During the last decade, because of DL algorithm success in various applications and according to Google Trends, DL have received more attention compared to other popular machine learning algorithms. In this regard, this research investigates recent advancements in DL algorithms and their applications in advanced analytic approaches for Big Data analyzing.

Keywords: Deep learning, Analytics, Big Data, Industry 4.0, Industry 5.0

References

[1] L. Da Xu, E.L. Xu, L. Li, Industry 4.0: State of the art and future trends, International Journal of Production Research. 56 (2018) 2941–2962. doi:10.1080/00207543.2018.1444806.

[2] B. Chen, J. Wan, L. Shu, P. Li, M. Mukherjee, B. Yin, Smart Factory of Industry 4.0: Key Technologies, Application Case, and Challenges, IEEE Access. 6 (2017) 6505–6519. doi:10.1109/ACCESS.2017.2783682.

[3] H. Ardakani, H. Hemmati, M.R. Hernandez, Data Driven Process Monitoring, from Fault Detection and Diagnosis Points of View, in Industry 4.0 Context, 32 European Symposium on Computer Aided Process Engineering, Elsevier, 2022: pp. 1033–1038. doi:https://doi.org/10.1016/B978-0-323-95879-0.50173-9.

[4] M. Chen, S. Mao, Y. Zhang, V.C.M. Leung, Big Data challenges and technologies, related future prospects, 2010. http://www.springer.com/series/10028.

[5] Q. Zhang, L.T. Yang, Z. Chen, P. Li, A survey on deep learning for big data, Information Fusion. 42 (2018) 146–157. doi:10.1016/j.inffus.2017.10.006.

[6] M.M. Najafabadi, F. Villanustre, T.M. Khoshgoftaar, N. Seliya, R. Wald, E. Muharemagic, Deep learning applications and challenges in big data analytics, Journal of Big Data. 2 (2015) 1–21. doi:10.1186/s40537-014-0007-7.

[7] X. Wang, Y. Zhao, F. Pourpanah, Recent advances in deep learning, International Journal of Machine Learning and Cybernetics. 11 (2020) 747–750. doi:10.1007/s13042-020-01096-5.

[8] B. Jan, H. Farman, M. Khan, M. Imran, I.U. Islam, A. Ahmad, et al., Deep learning in big data Analytics: A comparative study, Computers and Electrical Engineering. 75 (2019) 275–287. doi:10.1016/j.compeleceng.2017.12.009.