Double Descent - What is double descent phenomenon?

An interesting phenomenon in the field of machine learning is double descent. The test error goes down when the ratio of model parameters to data becomes up. Even when models are severely overparameterized, which happens when data is undersampled, this pattern persists.