Circumventing Outliers of AutoAugment with Knowledge Distillation

By Longhui Wei, Anxiang Xiao, Lingxi Xie, Xin Chen, Xiaopeng Zhang, Qi Tian. 2020

68 citations recorded by the source corpus.

AutoAugment has been a powerful algorithm that improves the accuracy of many vision tasks, yet it is sensitive to the operator space as well as hyper-parameters, and an improper setting may degenerate network optimization. This paper delves deep into the working mechanism, and reveals that AutoAugment may remove part of discriminative information from the training image and so insisting on the ground-truth label is no longer the best option. To relieve the inaccuracy of supervision, we make use of knowledge distillation that refers to the output of a teacher model to guide network training. Experiments are performed in standard image classification benchmarks, and demonstrate the effectiveness of our approach in suppressing noise of data augmentation and stabilizing training. Upon the cooperation of knowledge distillation and AutoAugment, we claim the new state-of-the-art on ImageNet classification with a top-1 accuracy of 85.8%.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Efficiency & Compression

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Soham De
2. Samuel L. Smith
3. Andrew Brock
4. Zhenguo Li
5. Hang Xu
6. Wei Zhang
7. Lanqing Hong
8. Quoc V. Le

In Computer Vision and Pattern Recognition

1. Xiaogang Wang
2. Luc Van Gool
3. Zhenguo Li
4. Dacheng Tao
5. Alan L. Yuille
6. Ming-Hsuan Yang
7. Hang Xu
8. Andrew Zisserman

In Efficiency & Compression

1. Hussam Lawen
2. Xinyue Cai
3. Asaf Noy
4. Hadi Pouransari
5. Ning Kang
6. Andrew Brock
7. Emanuel Ben-Baruch
8. Wei Zhang

Citations in the corpus

• Sharpness-Aware Minimization for Efficiently Improving Generalization (2020, neutral): “, 2020), KDforAA (Wei et al., 2020), TBMSL-Net (Zhang et al.”
• VEGA: Towards an End-to-End Configurable AutoML Pipeline (2020, methodology): “Auto Augment (Wei et al., 2020).”
• Solving ImageNet: a Unified Scheme for Training any Backbone to Top Results (2022, methodology): “[40] demonstrated that KD suppresses noise from data augmentations, and enables using stronger augmentations.”
• Characterizing signal propagation to close the performance gap in unnormalized ResNets (2021, background): “BatchNorm has become a core computational primitive in deep learning (Ioffe & Szegedy, 2015), and it is used in almost all state-of-the-art image classifiers (Tan & Le, 2019; Wei et al., 2020).”
• Extracurricular Learning: Knowledge Transfer Beyond Empirical Distribution (2020, methodology): “These methods exploit domain knowledge to transform training examples to generate more data [51, 33, 17], learn a data generation policy [15, 14, 29, 35, 71, 62], augment the”