On the Implicit Bias in Deep-Learning Algorithms
It addresses the fundamental problem of understanding generalization in deep learning for researchers, but is incremental as a survey.
The paper surveys the concept of implicit bias in deep-learning algorithms, which is believed to explain their generalization ability despite overparameterization, by reviewing key results and discussing implications.
Gradient-based deep-learning algorithms exhibit remarkable performance in practice, but it is not well-understood why they are able to generalize despite having more parameters than training examples. It is believed that implicit bias is a key factor in their ability to generalize, and hence it was widely studied in recent years. In this short survey, we explain the notion of implicit bias, review main results and discuss their implications.