Search papers, labs, and topics across Lattice.
New York University, University of Amsterdam
4
0
6
19
Bi-encoder performance hinges on model variant, while cross-encoders maintain a robust edge by jointly encoding record pairs, revealing critical insights for entity matching strategies.
Uncertainty estimation in clinical VLMs fails to signal when models are unreliable, but it can predict which predictions are likely to fail under stress.
LLMs can significantly boost the utility of differentially private de-identification for clinical text, offering a path to better privacy-preserving data sharing.
Instruction-tuned MLLMs, despite excelling in zero-shot scenarios, surprisingly fail to leverage few-shot examples or chain-of-thought prompting, suggesting a fundamental limitation in their ability to learn from demonstrations.
