Search code examples
deep-learningmnistmatconvnetbatch-normalization

Does work Batch normalization when Batchnormalization=false to true in Matconvnet


I am using matconvnet library. In the MNIST example the default parameter batchnormalization=true. But VL_NNBNORM() function has nothing(just comment lines). Is it working though this? or Am i have to change vl_nnbnorm function? I mean that if i change batchnormalization=false to true, is it enough? Thanks..


Solution

  • Yes it's enough. If you see faster convergence, it means batch norm is working. vl_nnbnorm.m has nothing because it calls a mex file. This is how Matlab calls a mex file --- you have a dummy .m file with the same name.