Towards Modeling Uncertainties of Self-explaining Neural Networks via Conformal Prediction

• DOI: 10.48550/arxiv.2401.01549
• Publication Date: 2024-01-01
• Authors: Wei Qian; Chenxu Zhao; Yangyi Li; Fenglong Ma; Chengqi Zhang; Mengdi Huai
• Keywords: Interpretability; Deep Neural Networks

Despite the recent progress in deep neural networks (DNNs), it remains challenging to explain predictions made by DNNs. Existing explanation methods for DNNs mainly focus on post-hoc explanations where another explanatory model is employed provide explanations. The fact that can fail reveal actual original reasoning process of raises need build with built-in interpretability. Motivated this, many self-explaining have been proposed generate not only accurate but also clear and intuitive insights into why a particular decision was made. However, existing are limited providing distribution-free uncertainty quantification two simultaneously generated prediction outcomes (i.e., sample's final its corresponding interpreting prediction). Importantly, they establish connection between confidence values assigned interpretation layer those allocated ultimate layer. To tackle aforementioned challenges, this paper, we design novel modeling framework networks, which demonstrates strong performance excels producing efficient effective sets based informative high-level basis We perform theoretical analysis framework. Extensive experimental evaluation effectiveness