Post-Selection in Deep Learning is statistically invalid and traditional cross-validation does not rescue it.
Post-Selection in machine learning is statistically invalid, even with cross-validation.