Search papers, labs, and topics across Lattice.
West Virginia University
4
0
9
ReXTrust outshines all other methods, achieving a peak AUC of 93.0, revealing the critical importance of white-box access in hallucination detection for VLMs in clinical applications.
Class-conditioned perturbations reveal that in-distribution samples remain stable while OOD samples diverge, offering a new lens on model uncertainty.
BSTabDiff achieves superior synthetic data generation in HDLSS contexts by intelligently leveraging block-subunit structures to capture complex dependencies.
Forget reward hacking and entropy collapse: multi-reward RLIF, combining answer-level and completion-level signals, unlocks stable and robust LLM reasoning without human supervision.
