Milestones

List view

• (In the same vein) reproduce figure 3 with sigmoidal learning network

  Overdue by 6 year(s)

  •

  Due by June 17, 2020
  0% complete0 open0 closed

• Reproducing figure 3

  Once the setup for multiple runs is ready, a step toward reproducing Mahmood and Sutton (2013) work will be to focus on figure 3, that is, implementing the representation search methods.

  Overdue by 6 year(s)

  •

  Due by May 27, 2020
  0% complete0 open0 closed

• Reproduce figure 2 with sigmoid activation for the learning network

  Once the setup for multiple runs is ready, a good step up will be changing the activation for the learning network to sigmoids, keeping everything else the same, including the LTUs in the target network. Our conjecture is that figure 2 will be faithfully reproduced even with sigmoidal networks. It is a crucial step up because then we can easily apply backprop for all the layers of the learning network.

  Overdue by 6 year(s)

  •

  Due by June 3, 2020
  0% complete0 open0 closed