Smoothness Adaptive Hypothesis Transfer Learning