Skip to content

Muppet: A Modular and Constructive Decomposition for Perturbation-based Explanation Methods

The topic of explainable AI has recently received attention driven by a growing awareness of the need for transparent and accountable AI. In this paper, we focus on local perturbation-based explanation methods, which currently represent a majority of the post-hoc explainability literature and usage. We have observed a fundamental commonality among these methods, forming the essence of our contribution. We propose a novel methodology to decompose any state-of-the-art perturbation-based explainability approach into four blocks. This decomposition framework offers a concise analysis of existing

similarities between methods and accelerates the development of new ones and their variants by promoting the reuse of their common blocks. In addition, we provide Muppet: an open-source Python library that offers (i) explainable AI methods to debug and interpret ML models, (ii) standardized API based on our decomposition methodology, facilitating contribution and benchmarking, and (iii) built-in modularity design: enables the decomposition of every method into reusable modules, thus speeding up their implementation. Available soon at https://github.com/euranova/muppet/.

Quentin Ferre, Ismail Bachchar, Hakima Arroubat, Aziz Jedidi, Youssef Achenchabe, Antoine Bonnefoy, Muppet: A Modular and Constructive Decomposition for Perturbation-based Explanation Methods, In Proc. of the International Joint Conference on Neural Networks, July 2025.

Click here to access the paper. 

Releated Posts

Insights from GTC Paris 2025

Among the NVIDIA GTC Paris crowd was our CTO Sabri Skhiri, and from quantum computing breakthroughs to the full-stack AI advancements powering industrial digital twins and robotics, there is a lot to share! Explore with Sabri GTC 2025 trends, keynotes, and what it means for businesses looking to innovate.
Read More

Development & Evaluation of Automated Tumour Monitoring by Image Registration Based on 3D (PET/CT) Images

Tumor tracking in PET/CT is essential for monitoring cancer progression and guiding treatment strategies. Traditionally, nuclear physicians manually track tumors, focusing on the five largest ones (PERCIST criteria), which is both time-consuming and imprecise. Automated tumor tracking can allow matching of the numerous metastatic lesions across scans, enhancing tumor change monitoring.
Read More