Ask, Attend, Attack: An Effective Decision-Based Black-Box Targeted Attack for Image-to-Text Models

Keyphrases

• Black Box 100%
• Text-to-image 100%
• Decision Basis 100%
• Targeted Attack 100%
• Text Model 100%
• Adversarial Attack 33%
• Target Text 33%
• Attacker 33%
• Semantic Loss 33%
• Optimization Problem 16%
• Transformer-based 16%
• Search Space 16%
• Evolutionary Algorithms 16%
• Training Process 16%
• Semantically Related 16%
• Three-stage Process 16%
• Large-scale Optimization Problems 16%
• White-box Attack 16%
• CNN-RNN 16%
• Vision-language Tasks 16%
• Attack Performance 16%
• Gray Hole Attack 16%
• CNN-Transformer 16%

Computer Science

• Targeted Attack 100%
• Optimization Problem 33%
• Adversarial Machine Learning 33%
• Attackers 33%
• Experimental Result 16%
• Convolutional Neural Network 16%
• Search Space 16%
• Recurrent Neural Network 16%
• Evolutionary Algorithm 16%
• Challenging Scenario 16%
• Training Process 16%
• Large-Scale Optimization 16%