arxiv: 2605.08858 · v1 · submitted 2026-05-09 · 💻 cs.CV

Recognition: no theorem link

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

Piotr Borycki , Magdalena Tr\k{e}dowicz , Jacek Tabor , {\L}ukasz Struski , Przemys{\l}aw Spurek

Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:19 UTC · model grok-4.3

classification 💻 cs.CV

keywords post-hoc explainabilityprototype-based explanationsdata-free methodsgenerative modelsXAIneural network interpretabilityprivacy-preserving AIcomputer vision

0 comments

The pith

ProDG generates high-fidelity visual prototypes for explaining neural network decisions using only the model's weights and no real data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces ProDG, a framework that applies generative models to create prototypes for post-hoc explanations of model predictions. Earlier prototype-based methods needed access to some data samples to locate suitable prototypes, creating a barrier in restricted settings. By pulling prototypes straight from the frozen network weights, the approach removes that requirement entirely. A reader would care because it makes intuitive 'this looks like that' explanations feasible in privacy-sensitive or data-inaccessible environments.

Core claim

ProDG leverages generative models to synthesize pure, high-fidelity prototypes directly from the frozen model's weights, completely eliminating the dependency on any external data for prototype-based post-hoc explainability.

What carries the argument

Generative models that synthesize prototypes directly from the frozen model's weights to replace the data-dependent search step in prototype selection.

Load-bearing premise

Synthetic prototypes produced from the model weights will match the visual and semantic properties that real data prototypes would have for the same model decisions.

What would settle it

Apply both ProDG and a data-based prototype method to the same pretrained image classifier on a public dataset, then compare whether the resulting prototypes produce equivalent nearest-prototype classification accuracy and human-rated explanation faithfulness.

Figures

Figures reproduced from arXiv: 2605.08858 by Jacek Tabor, {\L}ukasz Struski, Magdalena Tr\k{e}dowicz, Piotr Borycki, Przemys{\l}aw Spurek.

**Figure 1.** Figure 1: Overview of the ProDG prototype retrieval pipeline and explanation generation. Given an input image, ProDG first applies a classification model to predict the target class. It then identifies the top-k most influential channels with respect to the predicted class. From an optimized prompt bank, ProDG samples prompts to generate corresponding images, which are subsequently used to compute activation heatmap… view at source ↗

**Figure 2.** Figure 2: Qualitative comparison of explanations produced by ProDG, Grad-CAM, and LRP ProDG reveals semantically rich visual cues, capturing object structure, color patterns, texture details, and discriminative regions such as the tortoise shell. In contrast, Grad-CAM and LRP primarily highlight coarse activation regions, which are less informative for interpreting specific visual concepts and attributes. prominent … view at source ↗

**Figure 3.** Figure 3: Our framework ProDG performs concept prompts optimization and feature disentanglement within the FLUX generative model. The Concept Prompt Bank parameterizes the text embeddings of a frozen generative model (FLUX) to synthesize prototypical images that maximize concept purity. This prompt bank uses a reparameterization trick offsets to ensure a diverse distribution of generated images. The Orthogonal Feat… view at source ↗

**Figure 4.** Figure 4: Comparison of explanations between ProDG (Ours) EPIC and InfoDisent. This comparison highlights the data-independent nature of ProDG, in contrast to the data-dependent explanation mechanisms of EPIC and InfoDisent. The comparison is conducted on a representation learned on top of pretrained ResNet34 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Critical Difference diagram comparing user preferences for prototype-based explanation methods (ProDG, InfoDisent, and EPIC) in a three-way evaluation setting. The diagram is based on average ranks computed from participant responses. Statistical significance is assessed using the Bonferroni-Dunn test with α = 0.05. Methods connected by a horizontal line are not significantly different, while disconnecte… view at source ↗

**Figure 6.** Figure 6: Qualitative ablation study over loss components. Each column shows samples generated under different optimization objectives. The first column shows the full model optimized with L = −LU +λregLreg+λdivLdiv, while the second and third columns remove Lreg and LU , respectively. The full objective enforces both semantic alignment and diversity. Removing Lreg leads to weaker constraints on the prompt embedding… view at source ↗

**Figure 7.** Figure 7: User study instructions and guide. Illustrative guide presented prior to the user-study questionnaire, demonstrating the concept of a prototype and how to interpret prototype-based explanations. The figure provides an intuitive guide for understanding visualizations in our framework [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Example questions (i) and (ii) from the first part of the user study. Participants were asked to evaluate the visual similarity between the generated prototype and the input image, as well as the visual coherence of the prototypes within a row, using a 1-5 Likert scale. As reported in Tab. 4, participants achieved significantly higher-than-chance accuracy on ImageNet. For the CUB-200-2011 dataset, we obser… view at source ↗

**Figure 9.** Figure 9: Example question (iii) from the first part of the user study. Participants evaluated whether the specific concepts highlighted by the generative prototypes can actually be observed in the original input image [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

**Figure 10.** Figure 10: Example question (iv) from the second part of the user study. Participants were presented with alternative sets of prototypes most influential for selecting different classes, including the correct one, and were asked to select the one that best explains the prediction. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗

**Figure 11.** Figure 11: Example question (v) from the second part of the user study. Participants were presented with alternative sets of prototypes found by different models (ProDG, EPIC and InfoDisent). They were asked select the one that best explains the given input image [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗

**Figure 12.** Figure 12: Example question (vi) from the second part of the user study. Participants chose which initialization of our method produces prototypes that best capture the defining features of the input image. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗

**Figure 13.** Figure 13: Example ProDG prototypes generated for the Stanford Dogs dataset. Each row highlights prototypes from a specific channel, focusing on different dog features such as ears, nose, and fur. Observe that the dogs’ breeds observed in prototypes are similar [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗

**Figure 14.** Figure 14: Example ProDG prototypes generated for the Stanford Cars dataset. Each row highlights prototypes from a specific channel, focusing on different part of vehicles. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_14.png] view at source ↗

**Figure 15.** Figure 15: Qualitative comparison of visual explanations generated by ProDG (Ours), EPIC, and InfoDisent across both CUB-200-2011 and ImageNet datasets. The baselines utilize localized image crops to highlight features, whereas ProDG synthesizes complete images to encapsulate the learned concepts 16 [PITH_FULL_IMAGE:figures/full_fig_p016_15.png] view at source ↗

**Figure 16.** Figure 16: Ablation study over loss configurations I. A qualitative comparison of models trained with different subsets of the objective terms {LU ,Lreg,Ldiv}, including all seven non-empty combinations. Only the full model (bottom) effectively preserves both diversity and structural integrity. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_16.png] view at source ↗

**Figure 17.** Figure 17: Ablation study over loss configurations II. A qualitative comparison of models trained with different subsets of the objective terms {LU ,Lreg,Ldiv}, including all seven non-empty combinations. Only the full model (bottom) effectively preserves both diversity and structural integrity. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_17.png] view at source ↗

read the original 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

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper claims to generate faithful prototypes for explanations straight from a frozen model's weights with no data access at all, but the abstract shows zero experiments or validation to support that the outputs actually explain anything.

read the letter

The central claim here is that generative models can synthesize high-fidelity prototypes directly from classifier weights, removing the last data requirement that prior post-hoc prototype methods still needed. That is the one thing worth noting: it targets a practical constraint in privacy-sensitive settings where even a validation subset is off-limits. The abstract does a clean job of contrasting this against recent work that still searches in test data, and the positioning feels straightforward without overclaiming prior results. The method description itself is presented as a new frontier in data-free XAI, which is fair as far as the stated goal goes. What the paper does well is identify a real remaining bottleneck and sketch a generative route around it. The soft spot is the complete absence of any evidence. No quantitative results, no visual examples, no comparison to data-dependent baselines, and no check on whether the synthesized images actually align with the concepts the model learned or just exploit weight-space artifacts. The weakest link is the untested assumption that activation-maximization-style generation from weights alone will produce prototypes whose visual and semantic properties match what real data would have supplied. Without that alignment, the explanations lose their grounding. This is the kind of paper that belongs in a reading group focused on XAI methods or generative approaches to interpretability. Someone already working on prototype networks or data-free constraints could pull useful details from the framework, but a general reader will wait for the experiments before treating the claim as solved. I would send it to peer review. The idea is specific enough and the limitation it attacks is concrete; referees can require the missing validation and ablations without the work being dismissed outright.

Referee Report

2 major / 1 minor

Summary. The paper introduces ProDG, a framework for data-free post-hoc explainability that employs generative models to synthesize high-fidelity visual prototypes directly from the weights of a frozen classifier, removing any need for training, validation, or test data to select or extract prototypes for 'this looks like that' style explanations.

Significance. If validated, the data-free property would be a meaningful advance for prototype-based XAI in privacy-restricted settings (e.g., medical or proprietary data), extending prior post-hoc prototype methods that still require data access. The manuscript receives credit for clearly identifying the data-dependency limitation in existing work and for proposing a generative synthesis route, but the significance is currently undercut by the complete absence of any empirical support.

major comments (2)

[Abstract and §3 (Method)] Abstract and §3 (Method): the central claim that generative synthesis 'directly from the frozen model's weights' produces prototypes whose visual and semantic content align with real-data selections is load-bearing, yet no optimization objective, latent-space constraint, or regularization term is specified that would anchor generations to the training distribution and prevent out-of-distribution or spurious outputs.
[§5 (Experiments)] Absence of §5 (Experiments) and all associated tables/figures: no quantitative results, ablation studies, fidelity metrics, or comparisons against data-dependent prototype baselines are reported, so there is no evidence that the synthesized prototypes yield faithful explanations for the original model's decisions.

minor comments (1)

[Abstract] The GitHub project page is referenced; including even preliminary qualitative examples or pseudocode in the main text would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the thorough review and constructive criticism. We respond to each major comment below.

read point-by-point responses

Referee: [Abstract and §3 (Method)] Abstract and §3 (Method): the central claim that generative synthesis 'directly from the frozen model's weights' produces prototypes whose visual and semantic content align with real-data selections is load-bearing, yet no optimization objective, latent-space constraint, or regularization term is specified that would anchor generations to the training distribution and prevent out-of-distribution or spurious outputs.

Authors: We agree that the central claim requires a clear specification of the optimization process to ensure the generated prototypes align with the model's decision boundaries and the data distribution. The current manuscript describes the high-level approach but does not detail the objective function. We will update §3 with the full mathematical formulation, including the loss terms, latent constraints, and regularization to prevent out-of-distribution outputs. revision: yes
Referee: [§5 (Experiments)] Absence of §5 (Experiments) and all associated tables/figures: no quantitative results, ablation studies, fidelity metrics, or comparisons against data-dependent prototype baselines are reported, so there is no evidence that the synthesized prototypes yield faithful explanations for the original model's decisions.

Authors: We acknowledge that the current version of the manuscript does not include an experimental section, as it primarily presents the novel framework and its data-free approach. This is a valid concern regarding empirical support. In the revised manuscript, we will add a full §5 with experiments, including quantitative metrics for prototype fidelity, comparisons to data-dependent baselines, and ablation studies to validate that the generated prototypes provide faithful explanations. revision: yes

Circularity Check

0 steps flagged

No circularity: purely methodological framework introduction

full rationale

The paper describes a new framework (ProDG) for synthesizing prototypes via generative models from frozen classifier weights alone. No equations, derivations, fitted parameters, or predictions appear in the provided text. The central claim is the existence and utility of this data-free approach itself; it does not reduce any result to its own inputs by construction, nor rely on self-citation chains or imported uniqueness theorems. The description is self-contained as an independent methodological contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified premise that generative models can substitute for real-data prototype extraction while preserving explanation quality.

axioms (1)

domain assumption Generative models can synthesize prototypes whose explanatory value equals or exceeds that of prototypes extracted from real data samples.
The framework's value proposition depends on this equivalence holding for the target classifier.

pith-pipeline@v0.9.0 · 5504 in / 1117 out tokens · 45359 ms · 2026-05-12T01:19:33.162363+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

[1]

A unified approach to interpreting model predictions

Scott M Lundberg and Su-In Lee. A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30, 2017

work page 2017
[2]

why should i trust you?

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. "why should i trust you?" explaining the predictions of any classifier. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1135–1144, 2016

work page 2016
[3]

On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation.PLoS One, 10(7):e0130140, 2015

Sebastian Bach, Alexander Binder, Grégoire Montavon, Frederick Klauschen, Klaus-Robert Müller, and Wojciech Samek. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation.PLoS One, 10(7):e0130140, 2015

work page 2015
[4]

Grad-cam: Visual explanations from deep networks via gradient-based localization.International Journal of Computer Vision, 128:336–359, 2020

Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization.International Journal of Computer Vision, 128:336–359, 2020

work page 2020
[5]

This looks like that: deep learning for interpretable image recognition.Advances in neural information processing systems, 32, 2019

Chaofan Chen, Oscar Li, Daniel Tao, Alina Barnett, Cynthia Rudin, and Jonathan K Su. This looks like that: deep learning for interpretable image recognition.Advances in neural information processing systems, 32, 2019

work page 2019
[6]

Interpretable image classification with differentiable prototypes assignment

Dawid Rymarczyk, Łukasz Struski, Michał Górszczak, Koryna Lewandowska, Jacek Tabor, and Bartosz Zieli´nski. Interpretable image classification with differentiable prototypes assignment. InEuropean Conference on Computer Vision, pages 351–368. Springer, 2022

work page 2022
[7]

Protopshare: Prototypi- cal parts sharing for similarity discovery in interpretable image classification

Dawid Rymarczyk, Łukasz Struski, Jacek Tabor, and Bartosz Zieli´nski. Protopshare: Prototypi- cal parts sharing for similarity discovery in interpretable image classification. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 1420–1430, 2021

work page 2021
[8]

Neural prototype trees for interpretable fine-grained image recognition

Meike Nauta, Ron Van Bree, and Christin Seifert. Neural prototype trees for interpretable fine-grained image recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14933–14943, 2021

work page 2021
[9]

This looks like it rather than that: Protoknn for similarity-based classifiers

Yuki Ukai, Tsubasa Hirakawa, Takayoshi Yamashita, and Hironobu Fujiyoshi. This looks like it rather than that: Protoknn for similarity-based classifiers. InThe Eleventh International Conference on Learning Representations, 2022

work page 2022
[10]

Pip-net: Patch-based intuitive prototypes for interpretable image classification

Meike Nauta, Jörg Schlötterer, Maurice Van Keulen, and Christin Seifert. Pip-net: Patch-based intuitive prototypes for interpretable image classification. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2744–2753, 2023

work page 2023
[11]

Infodisent: Explainability of image classification models by information disentanglement.arXiv preprint arXiv:2409.10329, 2024

Łukasz Struski, Dawid Rymarczyk, and Jacek Tabor. Infodisent: Explainability of image classification models by information disentanglement.arXiv preprint arXiv:2409.10329, 2024

work page arXiv 2024
[12]

Side: Sparse information disentanglement for explainable artificial intelligence.arXiv preprint arXiv:2507.19321, 2025

Viktar Dubovik, Łukasz Struski, Jacek Tabor, and Dawid Rymarczyk. Side: Sparse information disentanglement for explainable artificial intelligence.arXiv preprint arXiv:2507.19321, 2025

work page arXiv 2025
[13]

Epic: Explanation of pretrained image classification networks via prototypes

Piotr Borycki, Magdalena Tr˛ edowicz, Szymon Janusz, Jacek Tabor, Przemysław Spurek, Arka- diusz Lewicki, and Łukasz Struski. Epic: Explanation of pretrained image classification networks via prototypes. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 17366–17373, 2026

work page 2026
[14]

Flux.https://github.com/black-forest-labs/flux, 2024

Black Forest Labs. Flux.https://github.com/black-forest-labs/flux, 2024. 10 l o o k s l i k e z o o m r e g i o n l o o k s l i k e l o o k s l i k e l o o k s l i k e prototype no. 1 prototype no. 2 prototype no. 3 prototype no. 4 Figure 7:User study instructions and guide.Illustrative guide presented prior to the user-study questionnaire, demonstrating ...

work page 2024