Skip to content

Optimal Importance Sampling for Diffusion processes applied to ISOKANN.

Notifications You must be signed in to change notification settings

axsk/OptImpSampling.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

95 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OptImpSampling.jl

A WIP collection of algorithms related to optimal importance sampling (OIS).

With a focus on ISOKANN:

  • ociso: OIS for Koopman eigen- and chi functions

  • isokann: ISOKANN using OIS for the chi approximations

  • control: simpler rewrite of ociso

  • isonew: rewrite of isokann, uses humboldtsampling

  • isokann2: sketch of ISOKANN in higher dimensions

Utils:

  • sqra: Square root approximation for the generators of Ito diffusions (copy from Sqra.jl)
  • humboldtsample: sample uniform along the reaction coordinate

Experimental implementations:

  • logvar: OIS of path functionals with control variates (implements Richter, Nusken 2021)
  • reinforce: OIS of functions via REINFORCE
  • reinforcepath: like reinfore but on path space (equivalent to logvar)
  • fbsde: Forward backward SDE with Neural Controls, learning all times at once (like Kebiri, Hartmann 2019)