Visualization of neural error functions

  • Objective: understanding convergence and visualizing neural network error functions.
  • Participants: Miroslaw Kordos, Wlodzislaw Duch
  • Time: since 2003
  • Results: papers, programs, talks:

    See papers with Mirek Kordos in this list.

    Error surfaces determine the efficiency of network training; they can be visualized in PCA space of the weigh sequences; surprisingly we have never seen local minima, only large plateaus.

  • Open problems: applications, theory.

    •  
  • Working Log (local access only)

    •