Results for "leakage detection"

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

Automated detection/prevention of disallowed outputs (toxicity, self-harm, illegal instruction, etc.).

AI focused on interpreting images/video: classification, detection, segmentation, tracking, and 3D understanding.

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

Recovering training data from gradients.

Extracting system prompts or hidden instructions.

Identifying and localizing objects in images, often with confidence scores and bounding rectangles.

Detects trigger phrases in audio streams.

Identifying abrupt changes in data generation.

Identifying suspicious transactions.