If I train on data set A first and the result of the training is that the variance of the model parameters is small, then if I infer on a completely different data set, will the variance of the data after the reasoning be small as well?
Could you rephrase your question? And, consider adding a concrete example—the current form seems to conflate concepts and is therefore hard to answer. 