The field of artificial intelligence is moving towards developing more robust and secure models, particularly in the context of adversarial attacks and data perturbations. Recent research has focused on enhancing model resistance to transferable adversarial examples, backdoor attacks, and input noise. Innovative approaches, such as trigger activation and proactive model robustification, have shown promising results in improving model security. Furthermore, the development of frameworks for evaluating model robustness, such as FLUKE and RoMA, has enabled more systematic and task-agnostic assessments of model vulnerability. Noteworthy papers in this area include: TrojanDam, which proposes a novel backdoor defense mechanism utilizing out-of-distribution samples, and Towards Model Resistant to Transferable Adversarial Examples via Trigger Activation, which introduces a new training paradigm for enhancing model robustness against transferable adversarial examples.
Advancements in Adversarial Robustness and Model Security
Sources
Towards Model Resistant to Transferable Adversarial Examples via Trigger Activation
TrojanDam: Detection-Free Backdoor Defense in Federated Learning through Proactive Model Robustification utilizing OOD Data
Impact of Noise on LLM-Models Performance in Abstraction and Reasoning Corpus (ARC) Tasks with Model Temperature Considerations
Property-Preserving Hashing for $\ell_1$-Distance Predicates: Applications to Countering Adversarial Input Attacks
FLUKE: A Linguistically-Driven and Task-Agnostic Framework for Robustness Evaluation
Evaluating the Vulnerability of ML-Based Ethereum Phishing Detectors to Single-Feature Adversarial Perturbations
Towards Robust LLMs: an Adversarial Robustness Measurement Framework