proteneer/timemachine

Gradient estimator precision

Opened this issue · 0 comments

Migrating from #654 (comment)


Some open discussion points:

  • How best to express gradient estimator precision? (variance per system parameter? per force field parameter? per parameter class?)
  • Can we avoid thinking about gradient estimator precision, and instead just think about estimator precision?
  • What options are available for improving the precision of these estimates, and how much do they help?