Browse Teach Curate

Multi-Head Explainer Framework for Saliency Interpretability in CNNs and Transformers

A modular framework that attaches to convolutional and Transformer networks to improve both accuracy and interpretability, built from an attention gate that highlights task-relevant features, deep supervision that guides early layers, and an equivalent-matrix module that fuses local and global representations into unified saliency maps. It integrates into architectures such as ResNet and BERT with minimal modification.

Rosie Text

Multi-Head Explainer: A General Scalable Framework to Improve Explainability in CNNs and

This paper introduces the Multi-Head Explainer (MHEX), a modular add-on that improves both the interpretability and the accuracy of convolutional neural networks and Transformer models. MHEX has thre…