Decision Based
Decision-based attacks target machine learning models by exploiting only their final output (decision), without requiring access to internal model parameters or gradients, posing a significant real-world security threat. Current research focuses on developing efficient algorithms, such as those employing evolutionary strategies, geometric approaches, and reinforcement learning, to generate adversarial examples that fool these models with minimal queries. These attacks are being explored across various model architectures, including convolutional neural networks, vision transformers, and graph neural networks, and their effectiveness highlights the need for robust defenses in diverse applications ranging from image recognition to semantic segmentation and beyond.