Section: New Results

Game theory and macromolecular modeling

Initially based on a very coarse representation of RNA, we refined our model of RNA folding as a game, using on-lattice coordinates and statistical potentials for the utility function. The resulting algorithm was implemented in the subsequent version of the GARN  [2] software.

The final year of Amélie Héliou's PhD led to theoretical developments in game theory, mainly obtained in collaboration with J. Cohen (LRI, Univ. Paris-Sud). First, the quasi-exponential convergence, under reasonable assumptions, of the HEDGE algorithm was demonstrated in collaboration with the POLARIS team in Grenoble [11]. Moreover, in a paper accepted at NIPS'17 [12], we addressed the learning of Nash equilibria. In this context, we established the convergence with high probability of no-regret learning in the bandit and semi-bandit settings.