Speaker: Georgios Piliouras, Singapore University of Technology and Design
We study both theoretically and experimentally a standard theoretical benchmark of game theory, non-atomicrouting games. Several learning dynamics are classically known to equilibrate in this setting due to its connectionto convex optimization. Nevertheless, as we will show if we keep upping the total demand in these systems, then they go through phase transitions and become formally chaotic. We provide both theoretical and experimental evidenceabout the robustness of this phenomenon. A new vocabulary emerges to describe this behavior, such as formal definitions of chaos. We will examine this new terminology and give several examples to make these notions more intuitive. Finally, we will also uncover new regularities in these chaotic systems. What type of order is hidden within chaos?