ProDG: Prototypes for Data-Free Generative Post-Hoc Explainability

2026-05-09

AI Summary

This work addresses a critical limitation of existing prototype-based post-hoc explainability methods, which rely on external data—such as test sets—and thus fail in scenarios where data is inaccessible or subject to privacy constraints. To overcome this, the paper introduces the first fully data-free generative prototyping framework that operates solely on frozen model weights, requiring neither original nor surrogate data. By optimizing directly in the latent space, the method generates high-fidelity, semantically pure visual prototypes that faithfully reflect model behavior. This approach establishes a novel paradigm termed “Data-Free Explainable AI” (Data-Free XAI), offering intuitive and reliable model interpretations while inherently preserving data privacy.

Abstract

Ante-hoc interpretability methods based on prototypes provide highly accurate explanations by utilizing the intuitive "this looks like that" reasoning paradigm. On the other hand, post-hoc models can explain predictions for a single image without relying on an underlying dataset or requiring costly neural network retraining. Recent approaches successfully solve the retraining problem for prototype-based networks. However, they still face a fundamental limitation: they require access to a subset of data (e.g., a test or validation set) to search for and extract the visual prototypes. In this paper, we address this issue and introduce ProDG: Generative Prototypes for Data-Free Post-Hoc Explainability, a novel framework that leverages generative models to synthesize pure, high-fidelity prototypes directly from the frozen model's weights, completely eliminating the dependency on any external data. By establishing this new frontier in Data-Free XAI, ProDG unlocks robust visual interpretability for privacy-sensitive domains, where original data is strictly restricted or fundamentally inaccessible. Project page: https://github.com/piotr310100/ProDG

Problem

• data-free explainability
• prototype-based explanation
• post-hoc interpretability
• visual prototypes
• privacy-sensitive domains

Innovation

• data-free explainability
• generative prototypes
• post-hoc interpretability
• prototype-based XAI
• model inversion