Improving LoRA with Variational Learning