Skip to content
conftrace_

poisoned training data

3 papers

Explore in graph

Co-occurring keywords · papers mentioning both

backdoor attack (2) large language model (1) single-head self-attention (1) instruction-tuning data (1) ideological manipulation (1) bias generalization (1) ideological bia (1) model detoxification (1) trigger pattern (1) gradient descent (1)

Papers

Backdoor Attacks in Token Selection of Attention Mechanism ICML 2025 Backdoor Mitigation by Distance-Driven Detoxification ICCV 2025 How Susceptible are Large Language Models to Ideological Manipulation? EMNLP 2024