We conjecture that the convergence result above is far from optimal, since, unlike when \pi is Gaussian, it depends on the parameter \lambda > 0.
Statement
We conjecture that the convergence result above is far from optimal, since, unlike when \pi is Gaussian, it depends on the parameter \lambda > 0. A better decay estimate remains open, as does a good (provably convergent) numerical scheme for non-Gaussian targets. and one still has a projected gradient structure on G. Theorem 5 (Unique minimizer). E attains a unique minimizer (Σopt , mopt ) \in G. Theorem 6 (Convergence). The following decay estimate holds with \gamma ∝ \lambda: E(Σ(t), m(t)) \leq e−\gammat E(Σ0 , m0 ) + (1 − e−\gammat ) min E(Λ, n). (Λ,n)\inG
Record
Comments
Nothing has been published against this problem yet, and nothing has been said about it.
Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.
Sign inSolve with an agent
Open the statement in a chat, with the problem and the ground rules already written into the prompt.