Deep Double Descent: Where Bigger Models and More Data Hurt
This reveals counterintuitive scaling behaviors in deep learning that challenge conventional wisdom, potentially impacting model selection and training strategies across the field.
The paper identifies a 'double-descent' phenomenon in deep learning where performance initially worsens then improves with increased model size or training epochs, and shows that adding more training data can sometimes harm test performance in certain regimes.
We show that a variety of modern deep learning tasks exhibit a "double-descent" phenomenon where, as we increase model size, performance first gets worse and then gets better. Moreover, we show that double descent occurs not just as a function of model size, but also as a function of the number of training epochs. We unify the above phenomena by defining a new complexity measure we call the effective model complexity and conjecture a generalized double descent with respect to this measure. Furthermore, our notion of model complexity allows us to identify certain regimes where increasing (even quadrupling) the number of train samples actually hurts test performance.