MARGIN: Margin-Aware Regularized Geometry for Imbalanced Vulnerability Detection

TL;DR

MARGIN introduces a geometry-aware framework for vulnerability detection that adaptively learns discriminative representations, effectively handling class imbalance and improving detection stability and accuracy.

Contribution

The paper proposes MARGIN, a novel metric-based approach using adaptive margin learning and hyperspherical modeling to address geometric distortions in imbalanced vulnerability datasets.

Findings

MARGIN outperforms strong baselines on public datasets.

It achieves significant improvements in classification and detection accuracy.

The approach enhances robustness, interpretability, and generalization.

Abstract

Software vulnerability detection is critical for ensuring software security and reliability. Despite recent advances in deep learning, real-world vulnerability datasets suffer from two severe challenges: frequency imbalance and difficulty imbalance. We reinterpret these challenges from an embedding geometry perspective, observing that such imbalances induce geometric distortions in hyperspherical representation space. To address this issue, we propose MARGIN, a metric-based framework that learns discriminative vulnerability representations through adaptive margin metric learning and hyperspherical prototype modeling. MARGIN dynamically adjusts geometric regularization according to the distribution structure estimated by the von Mises-Fisher concentration, aligning the probability mass of embedding distributions with their corresponding Voronoi cells, thereby reducing geometric distortion and yielding more stable decision boundaries. Extensive experiments on public vulnerability datasets show that MARGIN consistently outperforms strong baselines, achieving notable improvements in classification and detection, especially on challenging, imbalanced datasets. Further analysis demonstrates that MARGIN produces more structured embedding geometries, improving robustness, interpretability, and generalization.