VidModEx: Interpretable and Efficient Black Box Model Extraction for High-Dimensional Spaces

TL;DR

This paper introduces VidModEx, a novel black-box model extraction method for high-dimensional data that leverages SHAP for synthetic data generation, significantly improving accuracy in image and video classification tasks.

Contribution

The paper presents a new approach combining SHAP with energy-based GANs to enhance black-box model extraction in high-dimensional spaces, outperforming existing methods.

Findings

Achieved 16.45% accuracy improvement in image classification.

Extended to video classification with an average of 26.11% improvement.

Effective across various scenarios with different prediction information availability.

Abstract

In the domain of black-box model extraction, conventional methods reliant on soft labels or surrogate datasets struggle with scaling to high-dimensional input spaces and managing the complexity of an extensive array of interrelated classes. In this work, we present a novel approach that utilizes SHAP (SHapley Additive exPlanations) to enhance synthetic data generation. SHAP quantifies the individual contributions of each input feature towards the victim model's output, facilitating the optimization of an energy-based GAN towards a desirable output. This method significantly boosts performance, achieving a 16.45% increase in the accuracy of image classification models and extending to video classification models with an average improvement of 26.11% and a maximum of 33.36% on challenging datasets such as UCF11, UCF101, Kinetics 400, Kinetics 600, and Something-Something V2. We further demonstrate the effectiveness and practical utility of our method under various scenarios, including the availability of top-k prediction probabilities, top-k prediction labels, and top-1 labels.