Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via In-the-wild Cascading Flow Optimization

2025-02-04

AI Summary

To address the limited transferability of instance-agnostic generative methods in black-box multi-objective adversarial attacks, this paper proposes a dual-stream cascaded framework that decouples target-guided optimization from perturbation generation. It introduces an out-of-distribution cascaded training mechanism and implicit velocity field modeling to overcome generator capacity constraints. Additionally, a multi-objective gradient coupling strategy is designed to enable strong cross-model transferability. On the Inception-v3 → ResNet-152 transfer task, the method achieves a 34.58% improvement in attack success rate, while maintaining significant efficacy against robust models, including adversarially trained ones. This work presents the first instance-agnostic, generative black-box attack achieving high transferability across multiple objectives—marking a substantial advancement in practical adversarial threat modeling.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Adversarial attacks are widely used to evaluate model robustness, and in black-box scenarios, the transferability of these attacks becomes crucial. Existing generator-based attacks have excellent generalization and transferability due to their instance-agnostic nature. However, when training generators for multi-target tasks, the success rate of transfer attacks is relatively low due to the limitations of the model's capacity. To address these challenges, we propose a novel Dual-Flow framework for multi-target instance-agnostic adversarial attacks, utilizing Cascading Distribution Shift Training to develop an adversarial velocity function. Extensive experiments demonstrate that Dual-Flow significantly improves transferability over previous multi-target generative attacks. For example, it increases the success rate from Inception-v3 to ResNet-152 by 34.58%. Furthermore, our attack method, such as adversarially trained models, shows substantially stronger robustness against defense mechanisms.

Problem

• Improving transferability of adversarial attacks
• Multi-target instance-agnostic attack framework
• Enhancing robustness against defense mechanisms

Innovation

• Dual-Flow framework enhances transferability
• Cascading Distribution Shift Training utilized
• Adversarial velocity function improves robustness