An Empirical Analysis of the Impact of Data Augmentation on Knowledge Distillation

Abstract

Generalization Performance of Deep Learning models trained using Empirical Risk Minimization can be improved significantly by using Data Augmentation strategies such as simple transformations, or using Mixed Samples. We attempt to empirically analyze the impact of such strategies on the transfer of generalization between teacher and student models in a distillation setup. We observe that if a teacher is trained using any of the mixed sample augmentation strategies, such as MixUp or CutMix, the student model distilled from it is impaired in its generalization capabilities. We hypothesize that such strategies limit a model's capability to learn example-specific features, leading to a loss in quality of the supervision signal during distillation. We present a novel Class-Discrimination metric to quantitatively measure this dichotomy in performance and link it to the discriminative capacity induced by the different strategies on a network's latent space.

References in corpus (17)

• Distilling the Knowledge in a Neural Network
• mixup: Beyond Empirical Risk Minimization
• Deep Residual Learning for Image Recognition
• Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
• Improved Regularization of Convolutional Neural Networks with Cutout
• On Calibration of Modern Neural Networks
• Weight Uncertainty in Neural Networks
• When Does Label Smoothing Help?
• DropBlock: A regularization method for convolutional networks
• Born Again Neural Networks
• Robustness May Be at Odds with Accuracy
• Towards Understanding Knowledge Distillation
• Data augmentation instead of explicit regularization
• Implicit Regularization in Deep Learning
• Affinity and Diversity: Quantifying Mechanisms of Data Augmentation
• Data Augmentation Revisited: Rethinking the Distribution Gap between Clean and Augmented Data
• On the convergence of mirror descent beyond stochastic convex programming

Cited by in corpus (4)

• Distilling Large Language Models into Tiny and Effective Students using pQRNN
• Robustness-Reinforced Knowledge Distillation with Correlation Distance and Network Pruning
• Leveraging Angular Distributions for Improved Knowledge Distillation
• Isotonic Data Augmentation for Knowledge Distillation