LoRA fine-tuning is more robust than full fine-tuning against backdoor attacks but more vulnerable to untargeted label-flipping poisoning, based on BERT/GLUE experiments and NTK-information-geometry analysis.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Does Low Rank Adaptation Lead to Lower Robustness against Training-Time Attacks?
LoRA fine-tuning is more robust than full fine-tuning against backdoor attacks but more vulnerable to untargeted label-flipping poisoning, based on BERT/GLUE experiments and NTK-information-geometry analysis.