Many of these challenges have been partly addressed, but we feel can be more fully dealt with.

Generalization challenges:

  • developmental questions (i.e. how did the networks of interest get established in the first place?)
  • single cell dynamics: complex bursting, plateau potentials, etc. can be included, but do not have fully appropriate affect on the determination of connection weights
  • hormonal influences
  • glial networks

Simulation challenges:

  • various learning methods (e.g. batch learning, backprop, etc.) could be more completely integrated in current environments
  • scaling up simiulations (many hardware/software issues arise here, but they generally aren't specific to the NEF)