Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Autoencoder based approach for the mitigation of spurious correlations

Srinitish Srinivasan, Karthik Seemakurthy

TL;DR

This paper proposes an autoencoder-based approach to analyze the nature of spurious correlations that exist in the Global Wheat Head Detection (GWHD) 2021 dataset and shows that this approach has the ability to suppress some of the spurious correlations in the GWHD 2021 dataset.

Abstract

Deep neural networks (DNNs) have exhibited remarkable performance across various tasks, yet their susceptibility to spurious correlations poses a significant challenge for out-of-distribution (OOD) generalization. Spurious correlations refer to erroneous associations in data that do not reflect true underlying relationships but are instead artifacts of dataset characteristics or biases. These correlations can lead DNNs to learn patterns that are not robust across diverse datasets or real-world scenarios, hampering their ability to generalize beyond training data. In this paper, we propose an autoencoder-based approach to analyze the nature of spurious correlations that exist in the Global Wheat Head Detection (GWHD) 2021 dataset. We then use inpainting followed by Weighted Boxes Fusion (WBF) to achieve a 2% increase in the Average Domain Accuracy (ADA) over the YOLOv5 baseline and consistently show that our approach has the ability to suppress some of the spurious correlations in the GWHD 2021 dataset. The key advantage of our approach is that it is more suitable in scenarios where there is limited scope to adapt or fine-tune the trained model in unseen test environments.

Autoencoder based approach for the mitigation of spurious correlations

TL;DR

This paper proposes an autoencoder-based approach to analyze the nature of spurious correlations that exist in the Global Wheat Head Detection (GWHD) 2021 dataset and shows that this approach has the ability to suppress some of the spurious correlations in the GWHD 2021 dataset.

Abstract

Deep neural networks (DNNs) have exhibited remarkable performance across various tasks, yet their susceptibility to spurious correlations poses a significant challenge for out-of-distribution (OOD) generalization. Spurious correlations refer to erroneous associations in data that do not reflect true underlying relationships but are instead artifacts of dataset characteristics or biases. These correlations can lead DNNs to learn patterns that are not robust across diverse datasets or real-world scenarios, hampering their ability to generalize beyond training data. In this paper, we propose an autoencoder-based approach to analyze the nature of spurious correlations that exist in the Global Wheat Head Detection (GWHD) 2021 dataset. We then use inpainting followed by Weighted Boxes Fusion (WBF) to achieve a 2% increase in the Average Domain Accuracy (ADA) over the YOLOv5 baseline and consistently show that our approach has the ability to suppress some of the spurious correlations in the GWHD 2021 dataset. The key advantage of our approach is that it is more suitable in scenarios where there is limited scope to adapt or fine-tune the trained model in unseen test environments.

Paper Structure

This paper contains 24 sections, 6 equations, 7 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Wheat Head Detection
  4. Spurious Correlations
  5. Proposed approach
  6. Preliminaries
  7. Explanation of Misclassified Instances using Autoencoders based on YOLOv5 Backbone
  8. Image Inpainting
  9. Glare removal using autoencoder
  10. Weighted Boxes Fusion
  11. Results and Discussion
  12. Hardware Environment
  13. Dataset
  14. Training Environment of Models used
  15. YOLOv5 Training Environment
  16. ...and 9 more sections

Figures (7)

  • Figure 1: Sample spurious correlations on GWHD 2021 dataset. (a) YoloV5 detections on the test sample, and (b) Sample image from training set.
  • Figure 2: False positives in the test split of GWHD 2021 dataset.
  • Figure 3: Architectural diagram of autoencoder based architecture to explain test set predictions.
  • Figure 4: Reconstructed Images of pixels activated for predicted wheat head spike regions in Fig. \ref{['fig:mis-outputs']}
  • Figure 5: Architecture Diagram of autoencoder for image glare correction.
  • ...and 2 more figures