Search papers, labs, and topics across Lattice.
1
0
3
0
VLMs forget less when learning new tasks if you explicitly preserve the semantic relationships they learned during pre-training, especially in vulnerable regions identified with adversarial anchors.
