Rashomon Sets for Prototypical-Part Networks: Editing Interpretable Models in Real-Time

2025-03-03

AI Summary

In high-risk visual tasks, ProtoPNet offers interpretability but suffers from an “interaction bottleneck”: model defect correction requires time-consuming retraining. This paper proposes Proto-RSet, the first framework to integrate the Rashomon set concept into prototype learning, enabling millisecond-scale interactive editing and debugging by non-expert users. Proto-RSet leverages Rashomon set sampling, constraint-based optimization, and differentiable prototype matching to rapidly generate multiple accurate and diverse ProtoPNet variants—while preserving performance (accuracy variation ≤ ±0.5%). We validate its efficacy on bias-mitigated bird recognition and clinical skin cancer diagnosis (debugging), with endorsement from domain experts. The core contribution is breaking the interaction bottleneck: Proto-RSet enables real-time, interpretable model correction without retraining.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Interpretability is critical for machine learning models in high-stakes settings because it allows users to verify the model's reasoning. In computer vision, prototypical part models (ProtoPNets) have become the dominant model type to meet this need. Users can easily identify flaws in ProtoPNets, but fixing problems in a ProtoPNet requires slow, difficult retraining that is not guaranteed to resolve the issue. This problem is called the"interaction bottleneck."We solve the interaction bottleneck for ProtoPNets by simultaneously finding many equally good ProtoPNets (i.e., a draw from a"Rashomon set"). We show that our framework - called Proto-RSet - quickly produces many accurate, diverse ProtoPNets, allowing users to correct problems in real time while maintaining performance guarantees with respect to the training set. We demonstrate the utility of this method in two settings: 1) removing synthetic bias introduced to a bird identification model and 2) debugging a skin cancer identification model. This tool empowers non-machine-learning experts, such as clinicians or domain experts, to quickly refine and correct machine learning models without repeated retraining by machine learning experts.

Problem

• Addresses the interaction bottleneck in ProtoPNets
• Enables real-time editing of interpretable machine learning models
• Facilitates quick model refinement by non-experts without retraining

Innovation

• Rashomon sets for diverse ProtoPNets generation
• Real-time editing of interpretable models
• Eliminates need for repeated retraining