Inconsistency, Instability, and Generalization Gap of Deep Neural Network Training
This work addresses the challenge of improving generalization in deep learning for practitioners, though it is incremental as it builds on existing concepts like sharpness and provides a basis for methods like co-distillation.
The paper tackles the problem of predicting and reducing the generalization gap in deep neural networks by analyzing inconsistency and instability of model outputs, showing that these metrics are strongly predictive of the gap and that reducing inconsistency leads to superior performance.
As deep neural networks are highly expressive, it is important to find solutions with small generalization gap (the difference between the performance on the training data and unseen data). Focusing on the stochastic nature of training, we first present a theoretical analysis in which the bound of generalization gap depends on what we call inconsistency and instability of model outputs, which can be estimated on unlabeled data. Our empirical study based on this analysis shows that instability and inconsistency are strongly predictive of generalization gap in various settings. In particular, our finding indicates that inconsistency is a more reliable indicator of generalization gap than the sharpness of the loss landscape. Furthermore, we show that algorithmic reduction of inconsistency leads to superior performance. The results also provide a theoretical basis for existing methods such as co-distillation and ensemble.