Data Science and Artificial Intelligence

Irena Atov, Kwang-Cheng Chen, Ahmed E. Kamal, Malamati Louta
2020 IEEE Communications Magazine  
F uture communication systems will be increasingly complex and heterogeneous, involving multiple networking technologies with different capabilities and characteristics and heterogeneous nodes with diverse features. All constituent elements will effectively interwork with the aim of advanced, high-quality service provisioning in a cost efficient manner, any time, any place in a seamless and transparent way, maintaining consistency, robustness/availability and service continuity. Diverse
more » ... ents should be satisfied, stringent performance metrics should be guaranteed, while systems should be enabled to adapt and efficiently evolve to ever changing conditions in a quick pace. Lately, the unprecedented amount of data availability in conjunction with the advancement in data analytics algorithms and computing processing have acted as a catalyst, allowing for the incorporation of artificial intelligence (AI) and machine learning (ML) capabilities into networks. In this way, knowledge acquisition and intelligent decision making support is enabled, considered to be a viable solution toward efficient network management, handling effectively increasing complexity, heterogeneity and highly dynamic networks' nature. AI/ML empowered future networks are enabled to sense their context of operation, analyze, reason and plan, make a decision, and act in accordance with the decision reached, while they learn from previous experience, thus optimizing their operation. Future networks are expected to have the ability to autonomously think, learn, remember, and adapt to changing conditions in order to achieve end-to-end goals and objectives. This Series is dedicated to the application of AI, ML, and data analytics to address different problems of communication systems, presenting new trends, approaches, methods, frameworks, systems for efficiently managing and optimizing networks related operations. Even though AI/ML is considered a key technology for next generation networks, still many research challenges need to be solved before it could reach its full potential. This Series has been increasingly popular, steadily receiving a greater number of submissions, despite its short lifetime. For this issue, only three months after the most recent one, eight articles were accepted following a rigorous review process by experts in the area in order to ensure the best possible papers were selected. The first six articles fall within future wireless systems (5G and beyond) design, while the last two apply AI/ML to promote security related solutions to defense mechanisms and end-to-end congestion control. The first article, "When Machine Learning Meets Wireless Cellular Networks: Deployment, Challenges, and Applications" by Ursula Challita, Henrik Ryden, Hugo Tullberg, provides an overview of the main requirements and key factors identified for efficiently deploying and integrating AI functionalities in 5G and beyond networks. To this respect, authors discuss the distribution of network intelligence, introducing three main types, as well as the challenges posed by a ML-based air interface supporting efficient data transmission, reducing energy consumption, while also satisfying latency requirements of different applications. Data acquisition, data security and integrity, and AI implementation are highlighted as key areas to be further investigated for a successful integration of AI in future wireless networks, while specific properties necessitated for AI-based systems include robustness and efficiency, AI goal alignment, active learning and explainable AI techniques. A diverse set of use case applications of AI to different networking problems are presented, including mobility management, wireless security, localization and physical layer, while the benefits that such techniques can bring to the network are highlighted. The authors conclude that an ML-based architecture for end-to-end communication system design along with an ML-based air interface are open research problems to be further investigated for initial deployments of AI-enabled wireless networks. In light of the aforementioned, the second article, "A Machine Learning-based Framework for Optimizing the Operation of Future Networks" by Claudio Fiandrino, Chaoyun Zhang, Paul Patras, Albert Banchs, and Joerg Widmer, proposes a general machine learning-based framework that leverages AI and ML tools to efficiently manage and optimize the performance of highly dynamic wireless networks. The proposed framework is modular and can instantiate and orchestrate multiple ML pipelines across different network segments for achieving different objectives. Machine intelligence is enabled into new as well as existing network functions, while reuse of existing control mechanisms with minimal or no modifications is succeeded. The authors use ML to forecast future traffic demands and characterize traffic features, advancing more intelligent decisions in critical network control mechanisms, such as load balancing, routing and scheduling. Their focus is on deep learning algorithms, while they additionally discuss the integration of their proposed solution in the 5G architecture. The proposed framework is validated, considering the proactive routing mechanism, and is shown to significantly reduce packet delay. In a similar line of work, identifying the importance of traffic prediction to the optimization of network resource management in future wireless systems, the third article, "Assisting for Intelligent Wireless Networks with Traffic Prediction: Exploring and Exploiting Predictive Causality in Wireless Traffic" by Juan Wen, Min Sheng, Jiandong Li, and Kaibin Huang, endeavors to improve traffic prediction accuracy by exploiting predictable causality which arises between occurrences of special events and triggered traffic variations. Traditional temporal and spatial correlation based prediction techniques mainly predict the regular component in traffic, constituting them ineffective for largescale varying traffic. Therefore, in order to tackle this limitation, the authors propose a novel framework of Correlation and Causality based Prediction (Coca-Predict) that integrates correlation and causality based prediction to exploit their complementary strengths, predicting regular component and variation tendency, hence maximizing the prediction accuracy. Experimental results
doi:10.1109/mcom.2020.9141187 fatcat:kszk7ystfbdo3bphkjunwr4ns4