Detecting and Filtering Unsafe Training Data via Data Attribution with Denoised Representation
Explore the Data Attribution with Denoised Representation (DRA) method, a novel approach for detecting and filtering unsafe training data to build more trust...