LoREnc secures foundation models and adapters by truncating dominant low-rank components and compensating only in authorized adapters, causing unauthorized outputs to collapse while authorized performance remains exact.
SOTER: guarding black-box inference for general neu- ral networks at the edge,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LoREnc: Low-Rank Encryption for Securing Foundation Models and LoRA Adapters
LoREnc secures foundation models and adapters by truncating dominant low-rank components and compensating only in authorized adapters, causing unauthorized outputs to collapse while authorized performance remains exact.