Search papers, labs, and topics across Lattice.
Yonsei University
1
0
3
Initializing prompts in flatter regions of the loss landscape dramatically improves calibration and performance in test-time prompt tuning for vision-language models.