Domaines
Statistical physics
Nonequilibrium statistical physics
Kinetic theory ; Diffusion ; Long-range interacting systems
Type of internship
Théorique, numérique Description
This internship aims at computing analytically the rate of convergence of the Heavy Ball method for quadratic and convex functions in the presence of stochastic resetting and possibly stochastic gradients. This analysis will show whether an optimal resetting rate may be chosen with minimal to no prior knowledge of the minimized function F. Lastly the dynamics of the Heavy Ball method with stochastic resetting will be characterised for functions F with rugged landscapes inspired by the potential energy of dense and disordered systems.
Contact
Yann-Edwin Keta