Komputation v0.6.2

  • Implemented adaptive learning rates through Adadelta, Adagrad and RMSProp
  • Added He initialization
  • Added an MNIST demo that uses He initialization, dropout and Nesterov’s Accelerated Gradient
  • Refactored the use of data set sizes in MnistData
  • Mentioned the new optimization strategies, He initialization and the new demo in the README