Investigating the influence of noise and distractors on the interpretation of neural networks