Adversarial Search
Adversarial search explores methods for finding inputs that intentionally mislead machine learning models, revealing vulnerabilities and improving robustness. Current research focuses on developing efficient attack algorithms, such as beam search methods, and robust defenses, including techniques leveraging uncertainty estimation and classical machine learning models for verification. This field is crucial for enhancing the security and reliability of AI systems across diverse applications, from image recognition and natural language processing to autonomous systems and search engines. The development of both effective attacks and defenses is driving improvements in model design and evaluation methodologies.
11papers
Papers
November 27, 2024
February 23, 2024
January 21, 2024
October 17, 2023
June 20, 2023
June 16, 2023
June 1, 2023