Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ctrl-A: Control-Driven Online Data Augmentation

Jesper B. Christensen, Ciaran Bench, Spencer A. Thomas, Hüsnü Aslan, David Balslev-Harder, Nadia A. S. Smith, Alessandra Manzin

Abstract

We introduce ControlAugment (Ctrl-A), an automated data augmentation algorithm for image-vision tasks, which incorporates principles from control theory for online adjustment of augmentation strength distributions during model training. Ctrl-A eliminates the need for initialization of individual augmentation strengths. Instead, augmentation strength distributions are dynamically, and individually, adapted during training based on a control-loop architecture and what we define as relative operation response curves. Using an operation-dependent update procedure provides Ctrl-A with the potential to suppress augmentation styles that negatively impact model performance, alleviating the need for manually engineering augmentation policies for new image-vision tasks. Experiments on the CIFAR-10, CIFAR-100, and SVHN-core benchmark datasets using the common WideResNet-28-10 architecture demonstrate that Ctrl-A is highly competitive with existing state-of-the-art data augmentation strategies.

Ctrl-A: Control-Driven Online Data Augmentation

Abstract

We introduce ControlAugment (Ctrl-A), an automated data augmentation algorithm for image-vision tasks, which incorporates principles from control theory for online adjustment of augmentation strength distributions during model training. Ctrl-A eliminates the need for initialization of individual augmentation strengths. Instead, augmentation strength distributions are dynamically, and individually, adapted during training based on a control-loop architecture and what we define as relative operation response curves. Using an operation-dependent update procedure provides Ctrl-A with the potential to suppress augmentation styles that negatively impact model performance, alleviating the need for manually engineering augmentation policies for new image-vision tasks. Experiments on the CIFAR-10, CIFAR-100, and SVHN-core benchmark datasets using the common WideResNet-28-10 architecture demonstrate that Ctrl-A is highly competitive with existing state-of-the-art data augmentation strategies.
Paper Structure (28 sections, 11 equations, 8 figures, 5 tables)

This paper contains 28 sections, 11 equations, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Offline methods
  4. Online methods
  5. ControlAugment
  6. Setting
  7. Concept and framework
  8. Updating the ASDs
  9. Determining $\boldsymbol{\Gamma}^{(j+1)}$
  10. Determining $\boldsymbol{\alpha}^{(j+1)}$
  11. Parameter control algorithm
  12. An extra role for the validation dataset
  13. ControlAugment in summary
  14. Experimental design and results
  15. Initial experiments with Ctrl-A
  16. ...and 13 more sections

Figures (8)

  • Figure 1: Illustration of the ControlAugment framework. Model training proceeds in phases of $n_p$ epochs, separated by the ControlAugment algorithm, which regulates the informed augmentation pool by adjusting the augmentation strength parameters $\boldsymbol{\Gamma}$ and $\boldsymbol{\alpha}$ for the subsequent training phase. The adjustment is executed by the ControlAugment block which quantifies the response of each operation using augmented validation data and through an internal control parameter ($\xi$) which is updated by comparing the setpoint value ($\kappa_{sp}$) with a relative training/validation performance metric ($\kappa$).
  • Figure 2: Examples of augmentation strength distributions with increasing distribution means from left to right.
  • Figure 3: Illustration of the Ctrl-A update procedure (Eqs. \ref{['eq:IArule']} and \ref{['eq:2ndIArule']}), which relies on relative operation response curves to determine new values for the ASD parameters $\Gamma$ and $\alpha$ based on the value of the control parameter, $\xi$. Implicitly, $\Gamma_3 = 1$ and $\alpha_1=\alpha_2 = 0$.
  • Figure 4: CIFAR-10 test accuracy (model: airbench-94) as a function of the control setpoint $\kappa_{sp}$ for (a) CtrlA($1$), (b) CtrlA($2$), and (c) CtrlA($3$). Search-optimized RA is represented by the shaded region ($95\,\%$ confidence interval).
  • Figure 5: Convergence results for CIFAR-10 performance.
  • ...and 3 more figures