Results for "target leakage"

Reusing knowledge from a source task/domain to improve learning on a target task/domain, typically via pretrained models.

Separating data into training (fit), validation (tune), and test (final estimate) to avoid leakage and optimism bias.

When information from evaluation data improperly influences training, inflating reported performance.

Recovering training data from gradients.

Extracting system prompts or hidden instructions.