Find hidden biases in vision AI without labeled bias data

Original: Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositions

Source: arxiv.org ↗