Deep Learning Dev. Kit Network Design
Network Design by User Interface
A software, dedicated to network design, is available. You can specify basic settings for your network, such as input data size. And also, you are allowed to design each layer without any programming.
Drop out setting, which is valid to avoid overfitting, is possible as well.
Multiple Termination Is Possible
You can create a network with multiple end outputs. Each output can have its own teaching data. It is useful to expand the availability of your network architectures.
Setting Elements on Network Layer
Layer Type
♦ Convolution
♦ Deconvolution
♦ Pooling
♦ Full Connection
♦ LSTM
♦ Transformation
♦ Connection To End
♦ Probability Per Pixel
♦ Pixelwise Convolution
Activation Function Type
♦ Sigmoid
♦ ReLu
♦ Softmax
♦ tanh
♦ Identity
There are other settings, such as drop out. We will develop new features on your request.
As Input for Other Systems
Created data can be used as an input for ezDeep, and you can see if your network is valid for an expected artificial intelligence.
Deep Learning Development Kit requires a network data which is created with this software.