Title: Is Deep Learning getting better?
Abstract: I will discuss one of the most unstudied but still most critical problems in deep learning, the lack of fair compatibility as the models progress. Due to this, we can’t claim models get better. I will base this discussion in some of the very few papers in the literature on this, and connect with issues like continual learning, robustness, and fairness. Will also discuss the lack of sub-group robustness in ML, yet another under-explored topic in deep learning.