-
Copyright-Protected Language Generation via Adaptive Model Fusion
An inference-time method for copyright protection that adaptively fuses models trained on disjoint copyrighted data.
-
Efficient randomized experiments using foundation models
A new estimator that integrates predictions from multiple foundation models into randomized experiments to reduce variance while preserving valid inference.
-
Semi-supervised novelty detection using ensembles with regularized disagreement
New state-of-the-art method for semi-supervised novelty detection with complex models like deep neural networks.