Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification

Suorong Yang, Furao Shen, Jian Zhao

TL;DR

EntAugment is proposed, a tuning-free and adaptive DA framework that outperform others without introducing any auxiliary models or noticeable extra computational costs, highlighting both effectiveness and efficiency.

Abstract

Data augmentation (DA) has been widely used to improve the generalization of deep neural networks. While existing DA methods have proven effective, they often rely on augmentation operations with random magnitudes to each sample. However, this approach can inadvertently introduce noise, induce distribution shifts, and increase the risk of overfitting. In this paper, we propose EntAugment, a tuning-free and adaptive DA framework. Unlike previous work, EntAugment dynamically assesses and adjusts the augmentation magnitudes for each sample during training, leveraging insights into both the inherent complexities of training samples and the evolving status of deep models. Specifically, in EntAugment, the magnitudes are determined by the information entropy derived from the probability distribution obtained by applying the softmax function to the model's output. In addition, to further enhance the efficacy of EntAugment, we introduce a novel entropy regularization term, EntLoss, which complements the EntAugment approach. Theoretical analysis further demonstrates that EntLoss, compared to traditional cross-entropy loss, achieves closer alignment between the model distributions and underlying dataset distributions. Moreover, EntAugment and EntLoss can be utilized separately or jointly. We conduct extensive experiments across multiple image classification tasks and network architectures with thorough comparisons of existing DA methods. Importantly, the proposed methods outperform others without introducing any auxiliary models or noticeable extra computational costs, highlighting both effectiveness and efficiency. Code is available at https://github.com/Jackbrocp/EntAugment.

EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification

TL;DR

EntAugment is proposed, a tuning-free and adaptive DA framework that outperform others without introducing any auxiliary models or noticeable extra computational costs, highlighting both effectiveness and efficiency.

Abstract

Data augmentation (DA) has been widely used to improve the generalization of deep neural networks. While existing DA methods have proven effective, they often rely on augmentation operations with random magnitudes to each sample. However, this approach can inadvertently introduce noise, induce distribution shifts, and increase the risk of overfitting. In this paper, we propose EntAugment, a tuning-free and adaptive DA framework. Unlike previous work, EntAugment dynamically assesses and adjusts the augmentation magnitudes for each sample during training, leveraging insights into both the inherent complexities of training samples and the evolving status of deep models. Specifically, in EntAugment, the magnitudes are determined by the information entropy derived from the probability distribution obtained by applying the softmax function to the model's output. In addition, to further enhance the efficacy of EntAugment, we introduce a novel entropy regularization term, EntLoss, which complements the EntAugment approach. Theoretical analysis further demonstrates that EntLoss, compared to traditional cross-entropy loss, achieves closer alignment between the model distributions and underlying dataset distributions. Moreover, EntAugment and EntLoss can be utilized separately or jointly. We conduct extensive experiments across multiple image classification tasks and network architectures with thorough comparisons of existing DA methods. Importantly, the proposed methods outperform others without introducing any auxiliary models or noticeable extra computational costs, highlighting both effectiveness and efficiency. Code is available at https://github.com/Jackbrocp/EntAugment.
Paper Structure (30 sections, 4 theorems, 10 equations, 6 figures, 9 tables, 1 algorithm)

This paper contains 30 sections, 4 theorems, 10 equations, 6 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data Augmentation
  4. Entropy-Regularized Loss
  5. The Proposed Method
  6. EntAugment
  7. Entropy-Regularized Loss
  8. Theoretical Analysis
  9. Complexity Analysis of EntLoss and EntAugment.
  10. Experiment
  11. Comparison with state-of-the-arts
  12. Implementation Details
  13. Results on CIFAR-10 and CIFAR-100
  14. Transfer Learning
  15. Results on Large-scale ImageNet
  16. ...and 15 more sections

Key Result

lemma thmcounterlemma

From the perspective of maximum likelihood estimation, the empirical risk minimizer can be denoted as $\hat{\boldsymbol{\theta}} = \arg \max_{\boldsymbol{\theta}} \mathbb{E}_{\mathbf{z} \sim \hat{p}_{\text{data }}} \log p_{\text{model }}(\boldsymbol{z}; \boldsymbol{\theta})$, where $\hat{p}_{data}$

Figures (6)

  • Figure 1: Training efficiency vs. classification accuracy. Training efficiency is calculated using the inverse of the average per epoch training time costs, where higher values are preferable. Regarding the efficiency-effectiveness trade-off, our method performs better than prior SOTA DA methods. EA+EL: EntAugment+EntLoss.
  • Figure 2: General EntAugment Procedure
  • Figure 3: Transferalbility analysis. Transferred test accuracy (%) on CIFAR-10.
  • Figure 4: t-SNE Visualization of CIFAR-10 dataset. DI: Dunn index. EA+EL: EntAugment+EntLoss.
  • Figure 5: Time consumption. Comparison of training cost of various DA methods.
  • ...and 1 more figures

Theorems & Definitions (6)

  • lemma thmcounterlemma
  • lemma thmcounterlemma
  • lemma thmcounterlemma
  • proposition thmcounterproposition
  • proof
  • proof