GHOST (Games, matHematical Optimization and STochastic systems) is a joint team between Inria, Université Grenoble Alpes and CNRS. We work at the interface of game theory, optimization, machine learning and stochastic dynamical systems.
Random dynamical systems are ubiquitous in many fields of computer science and applied mathematics: In machine learning for example, they are used to analyze learning algorithms and provide concrete convergence and generalization guarantees; in queuing theory, they are used to model, characterize, and optimize the performance of distributed systems; in game theory, they model the behavior of autonomous agents that are competing or cooperating to improve their performance; etc. Our team works at the interface of stochastic modeling, dynamical systems, online learning, game theory and optimization, and our aim is to (a) design mathematical and algorithmic methods for studying the dynamics of complex systems in the presence of randomness and uncertainty; and (b) to use these methods to optimize performance, design new learning algorithms and optimize decision-making in all its aspects. The applications of our work mostly revolve around complex resource allocation problems such as job allocation in distributed computing resources, improving the methods used to train complex machine learning models, and the areas of energy management and scheduling in electrical networks.