Devling into Adversarial Transferability on Image Classification: Review, Benchmark, and Evaluation

Researchers have conducted a comprehensive review of adversarial transferability in image classification, identifying gaps in standardized evaluation frameworks for transfer-based attacks. They propose a benchmark framework and categorize existing attacks into six distinct types to address biased assessments in current research.

• →Adversarial transferability allows attacks on models without direct access, creating significant security concerns.
• →Current evaluation methods for transfer-based attacks lack standardization, leading to potentially biased assessments.
• →Researchers reviewed hundreds of related works and organized transfer-based attacks into six distinct categories.
• →A comprehensive benchmark framework has been proposed to standardize evaluation of these attacks.
• →The study identifies common enhancement strategies and prevalent issues causing unfair comparisons in existing research.