Integrative CAM: Adaptive Layer Fusion for Comprehensive Interpretation of CNNs

Aniket K. Singh, Debasis Chaudhuri, Manish P. Singh, Samiran Chattopadhyay·December 02, 2024

Summary

Integrative CAM is an advanced technique for deep learning models, offering a comprehensive view of feature importance across CNNs. It addresses limitations of traditional gradient-based methods by fusing insights from all network layers, using both gradient and activation scores to adaptively weight layer contributions. This approach includes a novel bias term in saliency map calculation, essential for capturing feature importance. Integrative CAM generalizes the alpha term from Grad-CAM++ for wider model applicability. Through extensive experiments, it demonstrates superior fidelity in feature importance mapping, enhancing interpretability for complex decision-making tasks. This method provides a valuable tool for fusion-driven applications, promoting the deployment of deep learning models with trustworthiness and insight.

Key findings

11
  • header

Advanced features