Problem
Research questions and friction points this paper is trying to address.
long-tailed
adversarial training
class imbalance
robustness
decision boundary
Confusion-Geometry Rebalancing for Long-Tailed Adversarial Training
🤖 AI Summary
This work addresses the significant degradation in robustness for tail classes under long-tailed distributions, a problem exacerbated by class imbalance and perturbation amplification in adversarial training. To mitigate this, the authors propose Confusion-aware Geometric Rebalancing (CGRm), a plug-and-play method that introduces, for the first time, a directional confusion geometric graph to model inter-class relationships, identify critical decision boundaries, and enhance the robustness of vulnerable classes through feedback-weighted optimization and graph-guided boundary correction. By moving beyond the conventional paradigm of treating classes in isolation, CGRm consistently outperforms existing approaches across multiple long-tailed benchmarks. Ablation studies further confirm the effectiveness of each component, demonstrating a uniform improvement in robust performance.
📝 Abstract
Adversarial training under long tailed distributions suffers from a dual imbalance: the class imbalance skews the training objective toward head classes, and the adversarial inner maximization may further amplify this bias. Existing methods mitigate this issue by correcting class priors or adapting class wise robust supervision, yet they treat each class in isolation and fail to identify which boundaries drive long tailed collapse. We propose a Confusion Geometry Rebalancing method (CGRm) for long tail adversarial training, a plug in framework that leverages directed robust errors as training signals. CGRm leverages periodic robust evaluations to derive source class loss weights, class wise robust coefficients, and a directed confusion geometry graph. The method then couples feedback weighted robust optimization with graph guided margin correction, thereby boosting the robustness of vulnerable classes and sharpening the critical boundaries that drive long tailed performance degradation. Experiments on long tailed benchmarks show that CGRm achieves consistent robust performance gains over existing methods, with ablations validating the contribution of each component. We provide the code in the supplement.
Innovation
Methods, ideas, or system contributions that make the work stand out.
long-tailed adversarial training
confusion geometry
robust optimization
class imbalance
directed robust errors
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
