DL integration of prior knowledge

Deep Learning is not straightforward in general how to integrate prior knowledge into a deep learning system

Data hungry – Un/Semi/Weakly-supervised learning

This biases vision researchers to work on tasks where annotation is easy instead of tasks that are important

Generalization issue – Domain adaptation

Deep Nets perform well on benchmark datasets, but can fail badly on real world images outside the dataset

Robustness issues – Neural net layer design

Deep Nets are overly sensitive to changes in the image which would not fool a human observer