Search papers, labs, and topics across Lattice.
2
0
4
No existing model editing method consistently outperforms others in the complex landscape of medical VLMs, revealing significant trade-offs in reliability and generalizability.
Medical vision-language models perform better when the modality gap is tuned to an intermediate level, challenging the assumption that minimizing it is always optimal.
