Prof. Ali Sayed, EPFL

Title: Stochastic Diffusion Learning for Competing Networks with Time-varying Inference Strategies

Abstract: Stochastic diffusion learning provides an effective framework for modeling strategic interactions among competing networks. Two diffusion-based algorithms have recently been proposed for this setting: competitive diffusion (CD) learning, which extends adapt-then-combine diffusion from distributed optimization to network games, and adapt-then-combine and inference-combine (ATC-ITC), which further incorporates an inference step based on local information. Both algorithms have been shown to converge to an O(mu)-neighborhood of the Nash equilibrium under strongly monotone conditions. Although ATC-ITC achieves faster convergence than CD, it does not improve steady-state performance. This is because inference based on local information can enhance adversarial estimation during the transient phase, but may introduce persistent bias in steady state due to imperfect local knowledge of the adversarial network. To address this limitation, we introduce a time-varying step-size strategy for the inference learning step, which enables a trade-off between fast transient convergence and low steady-state mean-square error. We show that ATC-ITC equipped with this strategy can further reduce the magnitude of error terms introduced by the inference step. Simulation results validate the effectiveness of the proposed approach. Based on joint work with Y. Zhao, H. Cai.

Brief bio: A. H. Sayed leads the Adaptive Systems Laboratory at EPFL. He also served before as a distinguished professor and chair of electrical engineering at UCLA. He is a member of the US National Academy of Engineering and The World Academy of Sciences. He has also been recognized as a Highly Cited Researcher for several years and served as President of the IEEE Signal Processing Society in 2018 and 2019. He authored over 600 scholarly publications and nine books, including the 3-volume treatise on Inference and Learning from Data, published by Cambridge University Press in 2022. His research areas include adaptation and learning theories, data and network sciences, statistical inference, and multi-agent networks. His work has received several major awards including the 2022 IEEE Fourier Technical Achievement Award, the 2020 IEEE Wiener Society Award, the 2015 Education Award, the 2013 Meritorious Service Award, and the 2012 Technical Achievement Award from the IEEE Signal Processing Society, the 2014 Papoulis Education Award from the European Association for Signal Processing, the 2005 Terman Award from the American Society for Engineering Education, the 2003 Kuwait Prize, and the 1996 IEEE Donald G. Fink Prize. His group has been awarded several Best Paper Awards from the IEEE (2002, 2005, 2012, 2014) and EURASIP (2015). He is a Fellow of IEEE, EURASIP, and the American Association for the Advancement of Science (AAAS), the publisher of the journal Science.