Data-centric-approach-adressing-class-imbalance-and-overfitting-in-Convolutional-Neural-Network-CNN

The paradigm behind Deep Learning is now facing a shift from model-centric to data-centric. In this lab it is shown how data intricacies affect the outcome of a model. To show how far it will take to apply data changes without addressing the model, a single model is being used throughout a simple Convolutional Neural Network (CNN). While training this model the following common problems are addressed: class imbalance and overfitting. While facing these issues, the lab guides through useful diagnosis tools and methods to mitigate these common problems.