input perturbations
Deliberate alterations or variations introduced to the input data during training or testing to assess the robustness of AI models and their vulnerability to noise, ensuring models are resilient in real-world scenarios.
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization
- Defending Multimodal Backdoored Models by Repulsive Visual Prompt Tuning
- H-SPLID: HSIC-based Saliency Preserving Latent Information Decomposition
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization
- Learning (Approximately) Equivariant Networks via Constrained Optimization
- Learning (Approximately) Equivariant Networks via Constrained Optimization