| Full text | |
| Author(s): |
Bochie, Kaylani
[1]
;
Gilbert, Mateus S.
[1]
;
Gantert, Luana
[1]
;
Barbosa, Mariana S. M.
[1]
;
Medeiros, Dianne S. V.
[2]
;
Campista, Miguel Elias M.
[1]
Total Authors: 6
|
| Affiliation: | [1] Univ Fed Rio de Janeiro UFRJ, PEE COPPE DEL POLI, Grp Teleinformat & Automacao GTA, Rio De Janeiro, RJ - Brazil
[2] Univ Fed Fluminense UFF, PPGEET, MidiaCom, Niteroi, RJ - Brazil
Total Affiliations: 2
|
| Document type: | Journal article |
| Source: | JOURNAL OF NETWORK AND COMPUTER APPLICATIONS; v. 194, NOV 15 2021. |
| Web of Science Citations: | 0 |
| Abstract | |
Computer networks are dealing with growing complexity, given the ever-increasing volume of data produced by all sorts of network nodes. Performance improvements are a non-stop ambition and require tuning fine-grained details of the system operation. Analyzing such data deluge, however, is not straightforward and sometimes not supported by the system. There are often problems regarding scalability and the predisposition of the involved nodes to understand and transfer the data. This issue is at least partially circumvented by knowledge acquisition from past experiences, which is a characteristic of the herein called ``challenged networks{''}. The addition of intelligence in these scenarios is fundamental to extract linear and non-linear relationships from the data collected by multiple sources. This is undoubtedly an invitation to machine learning and, more particularly, to deep learning. This paper identifies five different challenged networks: IoT and sensor, mobile, industrial, and vehicular networks as typical scenarios that may have multiple and heterogeneous data sources and face obstacles concerning connectivity. As a consequence, deep learning solutions can contribute to system performance by adding intelligence and the ability to interpret data. We start the paper by providing an overview of deep learning, further explaining this approach's benefits over the cited scenarios. We propose a workflow based on our observations of deep learning applications over challenged networks, and based on it, we strive to survey the literature on deep-learning-based solutions at an application-oriented level using the PRISMA methodology. Afterward, we also discuss new deep learning techniques that show enormous potential for further improvements as well as transversal issues, such as security. Finally, we provide lessons learned raising trends linking all surveyed papers to deep learning approaches. We are confident that the proposed paper contributes to the state of the art and can be a piece of inspiration for beginners and also for enthusiasts on advanced networking research. (AU) | |
| FAPESP's process: | 15/24494-8 - Communications and processing of big data in cloud and fog computing |
| Grantee: | Nelson Luis Saldanha da Fonseca |
| Support Opportunities: | Research Projects - Thematic Grants |