Don't miss the mismatch : investigating the objective function mismatch for unsupervised representation learning