Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Feature Fusion for Improved Classification: Combining Dempster-Shafer Theory and Multiple CNN Architectures

Ayyub Alzahem, Wadii Boulila, Maha Driss, Anis Koubaa

TL;DR

The paper tackles uncertainty in deep learning predictions by fusing decisions from multiple pre-trained CNNs via Dempster-Shafer Theory. It introduces a four-phase DST-based ensemble pipeline: feature extraction from each model, mass function construction, DST-based mass fusion, and expected-utility-based prediction. On CIFAR-10 and CIFAR-100, the DST ensemble achieves accuracy improvements of 5.4% and 8.4% over the best single models, demonstrating the method's effectiveness. The results highlight DST as a robust framework for uncertainty-aware ensemble learning in practical DL deployments.

Abstract

Addressing uncertainty in Deep Learning (DL) is essential, as it enables the development of models that can make reliable predictions and informed decisions in complex, real-world environments where data may be incomplete or ambiguous. This paper introduces a novel algorithm leveraging Dempster-Shafer Theory (DST) to integrate multiple pre-trained models to form an ensemble capable of providing more reliable and enhanced classifications. The main steps of the proposed method include feature extraction, mass function calculation, fusion, and expected utility calculation. Several experiments have been conducted on CIFAR-10 and CIFAR-100 datasets, demonstrating superior classification accuracy of the proposed DST-based method, achieving improvements of 5.4% and 8.4%, respectively, compared to the best individual pre-trained models. Results highlight the potential of DST as a robust framework for managing uncertainties related to data when applying DL in real-world scenarios.

Feature Fusion for Improved Classification: Combining Dempster-Shafer Theory and Multiple CNN Architectures

TL;DR

The paper tackles uncertainty in deep learning predictions by fusing decisions from multiple pre-trained CNNs via Dempster-Shafer Theory. It introduces a four-phase DST-based ensemble pipeline: feature extraction from each model, mass function construction, DST-based mass fusion, and expected-utility-based prediction. On CIFAR-10 and CIFAR-100, the DST ensemble achieves accuracy improvements of 5.4% and 8.4% over the best single models, demonstrating the method's effectiveness. The results highlight DST as a robust framework for uncertainty-aware ensemble learning in practical DL deployments.

Abstract

Addressing uncertainty in Deep Learning (DL) is essential, as it enables the development of models that can make reliable predictions and informed decisions in complex, real-world environments where data may be incomplete or ambiguous. This paper introduces a novel algorithm leveraging Dempster-Shafer Theory (DST) to integrate multiple pre-trained models to form an ensemble capable of providing more reliable and enhanced classifications. The main steps of the proposed method include feature extraction, mass function calculation, fusion, and expected utility calculation. Several experiments have been conducted on CIFAR-10 and CIFAR-100 datasets, demonstrating superior classification accuracy of the proposed DST-based method, achieving improvements of 5.4% and 8.4%, respectively, compared to the best individual pre-trained models. Results highlight the potential of DST as a robust framework for managing uncertainties related to data when applying DL in real-world scenarios.
Paper Structure (18 sections, 11 equations, 3 figures, 4 tables)

This paper contains 18 sections, 11 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Frame of Discernment
  4. Basic Probability Assignment (BPA)
  5. Belief and Plausibility
  6. Doubt and Commonality
  7. Combination Rule
  8. Literature Review
  9. Proposed Algorithm
  10. Feature Extraction
  11. Mass Function Construction
  12. Mass Function Fusion
  13. Expected Utility Calculation and Prediction
  14. Evaluation and Results
  15. Analysis of Other Combinations
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Illustration of the Proposed Approach.
  • Figure 2: Prediction accuracy of pre-trained models and DST-based ensemble on CIFAR-10 and CIFAR-100. A zoomed-in section at the end of the distribution lines reveals the superiority of the DST-based ensemble in achieving higher accuracy.
  • Figure 3: Training progress of pre-trained models on CIFAR-10 and CIFAR-100. The subfigures indicate the validation accuracy over epochs, showcasing the comparative performance dynamics of the models.