RECAST: Model Reconstruction via Counterfactual-Aware Wasserstein Geometry under Limited Data

2026-06-26

AI Summary

This work addresses the challenge of high-fidelity model reconstruction in black-box settings where online queries are unavailable and data are scarce. Existing counterfactual explanation–based approaches suffer from decision boundary shift, overfitting, and reliance on strong assumptions. To overcome these limitations, we propose RECAST, the first method to integrate counterfactual-aware Wasserstein geometry into surrogate modeling. By leveraging Wasserstein barycenter prototypes, RECAST effectively incorporates counterfactual samples as informative yet atypical training data, enabling accurate reconstruction without requiring online access to the target model. The approach mitigates decision boundary shift, substantially improves reconstruction stability and query efficiency under low-data regimes, and facilitates systematic fairness auditing across groups. Empirical evaluations demonstrate that RECAST consistently achieves superior performance across multiple real-world datasets.

Abstract

Counterfactual explanations (CFs) help understand machine learning models by identifying minimal input changes that would lead to alternative model outcomes. Recent work demonstrates their utility for reconstructing black-box models, enabling third-party auditing of opaque decision systems for fairness and accountability. Still, CF-based reconstruction may suffer from decision boundary shifts, overfitting, and restrictive assumptions requiring online query access to target platforms. We propose REconstruction via Counterfactual-Aware waSserstein opTimization (RECAST) under limited data and restricted access, a behavioral surrogate model based on Wasserstein barycentric prototypes. Our approach addresses decision boundary shifts by incorporating CFs as informative, though less representative, samples for both classes, maintaining high surrogate fidelity in low-sample regimes without requiring online access during reconstruction. To enhance fairness auditing, our method enables systematic group fairness diagnostics. Experiments on real-world datasets and various setups show that RECAST effectively achieves high fidelity and query efficiency, as well as stable results even when the access is limited and noisy.

Problem

• counterfactual explanations
• model reconstruction
• limited data
• black-box models
• fairness auditing

Innovation

• Counterfactual Explanations
• Wasserstein Geometry
• Model Reconstruction
• Surrogate Modeling
• Fairness Auditing