machinelearning:: Models on the training set tend to exhibit double descent: as we get larger networks, training error will decrease, then increase, then decrease again.

Summary¶

Ideas¶

Abstract¶

Deep Double Descent (openai.com)