Shaoling Jing

text classification (1) harmful content (1) adversarial attack (1) perturbation robustness (1) safety evaluation (1) jailbreak attack (1) retrieval-augmented generation (1) syntax tree (1) llm safety (1) llm-generated text detection (1) syntactic feature (1) syntax tree feature (1) large language model (1) adversarial benchmark (1) red team testing (1) toxic content (1) llm detoxification (1) toxic subspace (1)