Skip to content

Milestones

List view

  • v0.2

    When we have: - a reasonably rich and polished feature map system - better organized network level of class hierarchy - PROBEN1 benchmark of our training method - (optional) revisited full doc

    Due by December 31, 2018
    •10/10 issues closed
    100% complete0 open10 closed

  • v0.1

    After finishing the specified milestone requirements, I think we reached a pretty well-rounded basic state with the library. Therefore I want to make our first release :)

    Due by July 15, 2018
    •7/7 issues closed
    100% complete0 open7 closed

  • NN Training and Variational Parameter System

    1) Own user-friendly least-squares fit training system, including training/validation/testing data split, early stopping technique and inclusion of regularization and derivative residuals. 2) Variational Parameter system allows to make a difference between betas / other parameters and variational parameters. Only parameters which are assigned to variational parameters will be changed by optimizers like the trainers. Only for those parameters the network will calculate variational / cross derivatives.

    Due by July 7, 2018
    •6/6 issues closed
    100% complete0 open6 closed

  • Basic FFNN class structure

    1) Basic FFNN, including extensible feeder and activation function interface 2) Hierarchic class structure for units and layers, allowing for clean extension in future 3) Generalized system for storing internal configuration in files and restoring from it 4) Performance optimization of the hotspots

    Due by June 24, 2018
    •6/6 issues closed
    100% complete0 open6 closed