Search papers, labs, and topics across Lattice.
University of California
1
0
Restoring global logit comparability in vision-language models boosts zero-shot classification accuracy by 15% while maintaining robustness across diverse domains.
