Loading...
標記為「representation」的 2 篇文章
運用表徵工程與激活操控,於表徵層級操弄模型行為。
Probe internal model representations to identify exploitable features與develop representation-level attacks.