Adversarial Influence

Adversarial influence research explores how malicious actors can manipulate machine learning models, particularly in safety-critical applications like autonomous driving and large language models, to elicit undesirable outputs. Current research focuses on developing both sophisticated attack methods (e.g., adversarial influence maximization, data poisoning) and robust defenses (e.g., data curation, transfer learning), often within specific model architectures like graph neural networks and LLMs. Understanding and mitigating adversarial influence is crucial for ensuring the reliability and trustworthiness of AI systems across various domains, impacting both the security of AI and the broader societal acceptance of its applications.