Search papers, labs, and topics across Lattice.
4
0
5
PLAG transforms tabular anomaly detection by leveraging pseudo-labels to uncover fine-grained, localized anomalies, outperforming traditional methods that struggle with label scarcity.
Current forgery detection models fall apart on realistic surveillance footage, but SurFITR, a new dataset of subtly manipulated surveillance images, closes the gap.
Generalist anomaly detection gets a major boost: InCTRLv2 leverages dual semantic perspectives to achieve state-of-the-art performance across diverse datasets without retraining.
Domain skew in federated learning can be tamed by decoupling and calibrating domain-specific features, leading to more consistent and generalizable global models.
