rctn/DeepBoltzmannRN

Issues

• Biases at the visible layer not used?

  #14 opened 10 years ago by briancheung
  1

• Numerical gradient doesn't match computed gradient

  #13 opened 10 years ago by briancheung
  3

• Explore mean-field convergence rate

  #9 opened 10 years ago by JesseLivezey
  1

• Vectorize over bit-flips in pretraining

  #12 opened 10 years ago by JesseLivezey
  0

• Need to pretrain models

  #3 opened 10 years ago by JesseLivezey
  0

• Use mean-field estimates to seed next mean field update after parameter changes

  #8 opened 10 years ago by JesseLivezey
  0

• Use better mean-field estimates for upper layers in pretraining

  #11 opened 10 years ago by JesseLivezey
  0

• Bug in ExHidden?

  #10 opened 10 years ago by briancheung
  1

• Change Sample Hidden

  #7 opened 10 years ago by JesseLivezey
  0

• Vectorize MPF updates

  #6 opened 10 years ago by JesseLivezey
  0

• How to deal with continuous values for units?

  #5 opened 11 years ago by JesseLivezey
  0

• Write ExHidden (Expectation for Hidden Units)

  #4 opened 11 years ago by JesseLivezey
  0

• Energy function sign issue

  #2 opened 11 years ago by briancheung
  2

• Gradient descent on in mpfTrain

  #1 opened 11 years ago by briancheung
  1