Haris Gačanin received his Dipl.-Ing. degree in Electrical engineering from University of Sarajevo in 2000. In 2005 and 2008, respectively, he received MSc and PhD from Tohoku University in Japan. He worked at Tohoku University until 2010 as Assistant Professor and joined Alcatel-Lucent (now Nokia) in 2010, where he established research on data-driven analysis of communication systems at physical and media access layers. Currently, he is department head at Bell Labs and adjunct teaching professor at KU Leuven. His professional interests relate to research confluence between artificial intelligence and physical-layer communications to establish autonomous wireless systems. He has 200+ scientific publications (journals, conferences and patens) and invited/tutorial talks. He is senior member of IEEE and IEICE and recipient of IEICE Communication Systems Best Paper Award (joint 2014, 2015, 2017), The 2013 Alcatel-Lucent Award of Excellence, the 2012 KDDI Foundation Research Award, the 2009 KDDI Foundation Research Grant Award, the 2008 JSPS Postdoctoral Fellowships for Foreign Researchers, the 2005 Active Research Award in Radio Communications, 2005 Vehicular Technology Conference (VTC 2005-Fall) Student Paper Award from IEEE VTS Japan Chapter and the 2004 Institute of IEICE Society Young Researcher Award. He was awarded by Japanese Government (MEXT) Research Scholarship in 2002.
We are now several years into explosion of machine learning (ML) in wireless networks, used to enrich decision-making by finding structures in data - knowledge discovery - as means to describe the user behavior and network performance. With new designs of wireless networks, complexity and dynamicity rises, network resources are scattered, and diversity of network elements increases. Consider these examples with interesting challenges: 1) massive number of Internet-of-Things devices, sensors and actuators give rise to the problem of dynamic network planning; 2) broadband wireless leads to problems with real-time radio resource management; 3) ultra-reliable communications require support of real-time adjustments on latency and reliability in the orders of 99,99999%. For such designs artificial intelligence (AI) is expected to support high adaptability with respect to wireless environment and its services (e.g. virtual reality). This talk discusses a paradigm shift from contemporary data-driven wireless with ML toward autonomous wireless with knowledge management by AI. We explore motivation, opportunities and methodology to adopt training-free AI methods for self-organization of wireless systems. We point out specific properties of wireless environment and classify future directions on training-free vs training-based systems. We start from popular data-driven ML techniques and briefly elaborate their benefits and shortcomings for wireless application mentioned above. The focus is on reinforcement learning as a major (training-free) representative of AI. We briefly discuss learning principles of intelligent agent with problem of random exploration for wireless-specific environment. We discuss principles of self-organization by synthesizing reasoning and learning with knowledge management. Finally, we end with a case study using wireless AI prototype for self-deployment and self-optimization. The talk provokes new coming challenges and unveil interesting future directions across multi-disciplinary research areas.