Dense Cross-Connected Ensemble Convolutional Neural Networks for Enhanced Model Robustness

TL;DR

CNN robustness to input variations and adversarial attacks is a critical barrier in image recognition. The authors introduce Dense Cross-Connected Ensemble CNN (DCC-ECNN), which fuses DenseNet-style dense connectivity with cross-path ensemble learning via cross-connections among three DenseNet paths to enable extensive feature sharing. The work provides architectural details, analyzes the role of cross-connections, and demonstrates improved robustness and accuracy on CIFAR10-C and standard CIFAR-10/100 datasets, including stronger adversarial resilience. This approach offers a more robust and potentially deployment-friendly alternative for real-world image recognition tasks, particularly in safety-critical settings.

Abstract

The resilience of convolutional neural networks against input variations and adversarial attacks remains a significant challenge in image recognition tasks. Motivated by the need for more robust and reliable image recognition systems, we propose the Dense Cross-Connected Ensemble Convolutional Neural Network (DCC-ECNN). This novel architecture integrates the dense connectivity principle of DenseNet with the ensemble learning strategy, incorporating intermediate cross-connections between different DenseNet paths to facilitate extensive feature sharing and integration. The DCC-ECNN architecture leverages DenseNet's efficient parameter usage and depth while benefiting from the robustness of ensemble learning, ensuring a richer and more resilient feature representation.

Figures (1)

• Figure 1: Architecture of Dense Cross-Connected Ensemble CNN.