How to save a convnet checkpoint with batchNormalizationLayer

I found trainMean and trainVariance of the batchnormalizationLayer are both empty for convnet_checkpoints saved using trainNet and trainOption('CheckpointPath', 'somepath'). This brings the problem that when I try to calculated validation error from a saved checkpoint, it will give the error saying the network is not trained.

Is there some way to set the trainOption so that trainMean and trainVariance will be saved?