A Comprehensive Survey on Self-Interpretable Neural Networks

TL;DR

Self-Interpretable Neural Networks (SINNs) integrate interpretability directly into neural architectures, addressing the limitations of post-hoc explanations. The authors propose a five-category taxonomy—attribution-based, function-based, concept-based, prototype-based, and rule-based—plus hybrid variants to unify design principles across domains. The survey covers image, text, graph, and deep reinforcement learning applications, summarizes quantitative evaluation metrics, and discusses open challenges and future directions, including benchmarks and multi-modal extensions. A publicly available GitHub resource is provided to track ongoing progress in SINNs. Together, the work clarifies the design space for intrinsically interpretable models and guides future cross-domain research.

Abstract

Neural networks have achieved remarkable success across various fields. However, the lack of interpretability limits their practical use, particularly in critical decision-making scenarios. Post-hoc interpretability, which provides explanations for pre-trained models, is often at risk of robustness and fidelity. This has inspired a rising interest in self-interpretable neural networks, which inherently reveal the prediction rationale through the model structures. Although there exist surveys on post-hoc interpretability, a comprehensive and systematic survey of self-interpretable neural networks is still missing. To address this gap, we first collect and review existing works on self-interpretable neural networks and provide a structured summary of their methodologies from five key perspectives: attribution-based, function-based, concept-based, prototype-based, and rule-based self-interpretation. We also present concrete, visualized examples of model explanations and discuss their applicability across diverse scenarios, including image, text, graph data, and deep reinforcement learning. Additionally, we summarize existing evaluation metrics for self-interpretability and identify open challenges in this field, offering insights for future research. To support ongoing developments, we present a publicly accessible resource to track advancements in this domain: https://github.com/yangji721/Awesome-Self-Interpretable-Neural-Network.

Figures (6)

• Figure 1: Comparison between self-interpretation and post-hoc.
• Figure 2: Taxonomy of self-interpretable neural networks.
• Figure 3: Graphical illustrations of rule-based self-interpretation.
• Figure 4: A graphical illustration of hybrid self-interpretation methods that combine different self-interpretation components.
• Figure 5: Examples for text explanations.