Negative initial weights improve learning in recurrent neural networks.

Davor Pavisic, Jean-Philippe Draye, Roberto Teran, Gustavo Calderon, Guy Cheron, Gaetan Libert

Browse the full ESANN paper archive.