CIFAR

[sorenhartmann]

• [x] SGD
• [x] SGD with MAP estimatino
• [x] VI
• [x] VI Mixture prior
• [ ] MCMC
• [ ] GPU
  • [ ] SGD
  • [x] VI
  • [ ] MCMC
• [ ] SGD with optimizers
• [ ] Consider dropout for sdg... maybe as option instead of map?

[sorenhartmann]

High learning rate for VI because of bad prior?