Docent lecture

About a decade ago deep Learning made its first widely-noticed impact in a computer vision benchmark on image classification -ILSVRC-. The main method was to train a large convolutional network on large fully-labelled datasets for best classification accuracy. The field has come a long way since then and this standard paradigm has been applied to various fields of science and sectors of industry. In this talk, I will talk about different ways in which one can go beyond the standard paradigm. Particularly, I will talk about interpretability and reliability of deep networks. I will motivate these directions from the perspective of a few real-world applications of deep learning including breast cancer imaging .