What question did this study set out to answer?

This research aims to create a hybrid framework that enhances the interpretability of deep neural networks for critical applications.

May 9, 2026Open Access

Towards Transparent Deep Learning: A Hybrid Framework for Explainable Artificial Intelligence (XAI).

Key Points

This research aims to create a hybrid framework that enhances the interpretability of deep neural networks for critical applications.
Introduced HT-DNN, integrating local surrogate explanations and global concept-based reasoning.
Implemented an automated accountability auditing module to identify spurious correlations.
Conducted experiments using the CIFAR-10 benchmark dataset to evaluate the framework's effectiveness.
HT-DNN enhances explanation fidelity, improving interpretability without sacrificing predictive accuracy.
Experimental results indicate a significant performance increase on the CIFAR-10 dataset.

Abstract

While Deep Neural Networks (DNNs) achieve state-of-the-art performance across various domains, their inherent lack of interpretability hinders their deployment in high-stakes, safety-critical applications. To bridge this gap, this paper proposes HT-DNN, a novel hybrid Explainable Artificial Intelligence (XAI) framework. HT-DNN uniquely integrates local surrogate explanations for instance-level transparency, global concept-based reasoning for macro-level understanding, and an automated accountability auditing module to detect spurious correlations. Experimental evaluations on the CIFAR-10 benchmark dataset demonstrate that HT-DNN significantly improves explanation fidelity and interpretability without compromising the underlying model's predictive accuracy.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper