I found, that after every time I run my program (with simple multylayer perceptron) I get the same init values for weights and biases. So, the question is how to set the random seed for Initializer?
You can set the random seed with
mxnet::RandomSeed(42); This will seed all random generator in MXNet. https://mxnet.incubator.apache.org/doxygen/namespacemxnet.html#a5bf83454c8f41259eb9d471f59feda2b