CEAR: Certified Ensemble Adversarial Robustness in DNNs
Several recent research papers explore novel methods for enhancing the adversarial robustness of deep neural networks. These studies introduce techniques such as ensemble-based approaches combining empirical and certified defenses, the synergistic use of noise and bilateral filters, and a Bayesian framework to model adversarial uncertainty. Additionally, one paper proposes a new classifier that balances discriminability with robustness, while another focuses on adversarial purification methods capable of handling non-additive perturbations. AI
IMPACT These diverse approaches aim to improve the reliability and security of AI systems against malicious attacks, potentially enabling wider adoption in safety-critical applications.