Vertical markets and industries are paving the way for a large diversity of heterogeneous services, use cases, and applications in future 5G networks. It is currently common understanding that for 5G networks to be able to satisfy those needs, a flexible, adaptable, and programmable architecture based on network slicing is required. With this approach, 5G network can be logically separated into multiple virtual slice, each one tailored to serve a specific service. In this context, operators need the ability to automate their architecture configuration and monitoring processes to reduce their OPerational EXpenditure (OPEX), and more importantly to ensure that that the Quality-of-Service (QoS) and Quality-of-Experience (QoE) requirements of the offered services are not violated . The use of Artificial Intelligence (AI) techniques is emerging as a promising solution to achieve these goals and to replace complex and expensive human- dependent decision-making processes.
Amongst the many algorithms part of the AI family and to its branch Machine Learning, Reinforcement Learning (RL) and Artificial Neural Networks (ANNs) based schemes are becoming very popular in the context of future 5G networks. In the context of wireless networks, many researchers have proposed RL schemes to optimize the radio resource allocation , the inter-cell interference , or load balancing functions . In addition, ANNs have been proposed as a tool for two main applications : 1) prediction, inference, and big data analytics and 2) self-organizing network operations. However, new scalable and robust solutions, characterized by faster learning phase, and adapted to the mobile network management and orchestration are required.
The objective of this thesis is to design, develop, and assess a scalable architecture for learning how to optimally orchestrate a 5G network architectures, which doesn't require a priori information (e.g., data sets) and efficiently performs in a complex environment. To do so, the PhD student will have to overview existing hierarchical/distributed learning mechanisms and methods proposed to speed up and stabilize the learning process, and understand the tradeoffs between complexity, performance, and learning time.
The organization of the work during the PhD can proceed as follows: 1) State-of-the-art on RL and related function approximation techniques, and on 5G network orchestration and management solutions; 2) Based on this overview, develop a mathematical framework to model the resource management in 5G networks; 3) Propose a hierarchical/distributed solutions for learning optimal network management policy through RL schemes; 4) Investigate how to increase the framework reliability to deal with system changes or malicious agents; 5) Final evaluation in relevant 5G use cases .
The ideal candidate for this position has a strong background on wireless networks and expertise in mathematics. Also, good communication skills, both oral and written English, are required; french knowledge is not mandatory. Some backgrounds in machine/reinforcement learning, optimization, simulations, and scientific language is a plus. We expect also the candidate to have notions of research techniques (documenting and reporting, work organization, independent working, and creativity).
The PhD student will join the Broadband Wireless Systems Laboratory within CEA-Leti, Grenoble, France, and will be inscribed to University of Rennes I. The thesis will be carried out in collaboration with the University of Toronto. The duration of the PhD is 3 years, and it is expected to start during the last trimester of 2018.
CEA-Leti, one of the three advanced-research institutes within CEA Tech, is focusing on creating value and innovation through technology transfer to its industrial partners. It is specialized in nanotechnologies and their applications, from wireless devices and systems, to biology, healthcare and photonics. In 2017, CEA ranked second in the Reuters World’s Most Innovative Research Institutions. With a staff of more than 1,900, Leti is based in Grenoble, France, and has offices in Silicon Valley, California, and Tokyo.
The CEA-Leti Broadband Wireless Systems Laboratory is conducting cutting-edge research in wireless communications for broadband and 5G systems, including advanced channel coding and modulation, transceiver design, access control protocols, and radio and network resource management. Its activities cover a large spectrum, from the specification, simulation and characterization, to the design of both SW and HW components for wireless communications.
Potential candidates should send a resume and a motivation letter in PDF along with 2 references to the email addresses:
 ETSI ISG ENI, "Improved Operator Experience through Experiential Networked Intelligence (ENI)," White Paper No. 22, Oct.2017.
 R. S. Sutton and A. G. Barto, "Reinforcement learning: An introduction," Second Edition, in progress, MIT Press, Cambridge, MA, 2018.
 C. Bishop, "Pattern Recognition and Machine Learning (Information Science and Statistics)," 1st edn. 2006. corr. 2nd printing edn." Springer, New York (2007).
 I. Sorin Comsa, A. De Domenico, and D. Kténas, "QoS-Driven Scheduling in 5G Radio Access Networks - A Reinforcement Learning Approach," IEEE Globecom, Singapour, 2017.
 A. De Domenico and D. Kténas, "Reinforcement Learning for Interference-Aware Cell DTX in Heterogeneous Networks," IEEE Wireless Communications and Networking Conference (WCNC), Barcelona, 2018.
 P. Munoz et al., "Fuzzy rule-based reinforcement learning for load balancing techniques in enterprise lte femtocells," IEEE Transactions on Vehicular Technology, vol. 62, no. 5, pp. 1962-1973, Jun 2013.
 M. Chen, U. Challita, W. Saad, C. Yin, and M. Debbah, "Machine learning for wireless networks with artificial intelligence: A tutorial on neural networks," 2017, arXiv preprint arXiv:1710.02913.
 ETSI ISG ENI, "Experiential Networked Intelligence (ENI) ENI use cases," 2018.
(c) GdR 720 ISIS - CNRS - 2011-2018.