GradLock stores private images inside frozen neural-network weights during training, letting an attacker extract them from a released model even after quantization, pruning, or fine-tuning.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Don't Trust the AI Ecosystem: Analyzing Privacy Leakage in Compromised Open-Source Components
GradLock stores private images inside frozen neural-network weights during training, letting an attacker extract them from a released model even after quantization, pruning, or fine-tuning.