Rethinking the Test-Time Prompt Tuning Objective from the Perspective of Calibration
This work proposes a new objective that replaces the conventional EM loss by aligning the original-view prediction with a target distribution derived from augmented views via cross-entropy, while adversarially incorporating the entropy of the target distribution to capture sample-specific uncertainty.
Jungwon Choi, Hyeon-Gi Jang, Kibok Lee et al.
· 0 citations