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

arXiv: https://arxiv.org/abs/2006.03810 · HTML: https://mlsanity.com/paper/2006.03810/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 Jun 2020.

## Built on by (most cited first, 11 of 11)

- [Curriculum Temperature for Knowledge Distillation](https://mlsanity.com/paper/2211.16231/) (2022)
- [Meta Knowledge Distillation](https://mlsanity.com/paper/2202.07940/) (2022)
- [Distilling Large Language Models into Tiny and Effective Students using pQRNN](https://mlsanity.com/paper/2101.08890/) (2021)
- [Cross-View Consistency Regularisation for Knowledge Distillation](https://mlsanity.com/paper/2412.16493/) (2024)
- [Isotonic Data Augmentation for Knowledge Distillation](https://mlsanity.com/paper/2107.01412/) (2021)
- [Robustness-Reinforced Knowledge Distillation with Correlation Distance and Network Pruning](https://mlsanity.com/paper/2311.13934/) (2023)
- [Leveraging Angular Distributions for Improved Knowledge Distillation](https://mlsanity.com/paper/2302.14130/) (2023)
- [Distilling Calibrated Student from an Uncalibrated Teacher](https://mlsanity.com/paper/2302.11472/) (2023)
- [Mixed Sample Augmentation for Online Distillation](https://mlsanity.com/paper/2206.12370/) (2022)
- [Revisiting Data Augmentation in Model Compression: An Empirical and Comprehensive Study](https://mlsanity.com/paper/2305.13232/) (2023)
- [Diffusion-Assisted Distillation for Self-Supervised Graph Representation Learning with MLPs](https://mlsanity.com/paper/2510.04241/) (2025)

## Abstract (arXiv)

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.