DifAttack++: Query-Efficient Black-Box Adversarial Attack via Hierarchical Disentangled Feature Space in Cross–Domain

Summary

DifAttack++ is a score-based black-box adversarial attack that works in a hierarchical disentangled feature space. It splits an image's latent representation into an Adversarial Feature and a Visual Feature using a cross-domain autoencoder, then optimizes only the Adversarial Feature using query feedback from the victim model. The authors report higher Attack Success Rate and query efficiency than state-of-the-art methods, with comparable visual quality.

Topics

• Adversarial machine learning

Related items

• InfoImage watermarking robustness under MLLM-guided semantic editing: Benchmark, defense, and diagnosisSimilar attack · Elsevier Security Journals
• LowYour Robot's Safety Functions Already Work. What If the Input Lies?Similar attack · VicOne Blog
• InfoSpeech Tokenizers are Vulnerable: Transferable Semantic Attack and Robust TokenizerSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoAdversarial Retraining for Machine Learning-Based Man-in-the-Middle Attack Detection in Smart Grid Industrial Control SystemsSimilar attack · OpenAlex (peer-reviewed AI security)
• LowMeasuring Robustness and Efficiency in a Connectome-Constrained Fly Visual System Model: Linking an Adversarial and a Perceptual IllusionSimilar attack · Arxiv (cs.RO + cs.CV security)