A Universal Trade-off Between the Model Size, Test Loss, and Training Loss of Linear Predictors

Open in new window