research-ml

Research about ML Componentization and Experiment Realization Service.

###Please Read Issues!

Research Objectives

1. ML Componentization.
2. Specify detail XML format for ML Components.
3. Specify the algorithm which translates from XML to tensorflow object code.
4. Prototyping objectives above.

What we support

• Model

  • linear regression (regularization is supported)
  • logistic regression (regularization is supported)
  • softmax regression (regularization is supported)
  • neural network
  • convolution neural network
  • recurrent neural network

• Layer type

  • none(normal layer)
  • convolution
  • recurrent
  • lstm
  • gru

• Initializer

  • random uniform
  • random normal

• Activation function

  • relu

• Pooling

  • max pooling

• Padding

  • same
  • valid

• Optimizer

  • gradient descent
  • adadelta
  • adagrad
  • momentum
  • adam
  • ftrl
  • rmsprop