Pinpoint Counterfactuals: Reducing social bias in foundation models via localized counterfactual generation
This work addresses bias mitigation in foundation models for downstream AI applications, offering an incremental improvement over existing counterfactual methods by reducing artifacts.
The paper tackled the problem of societal bias propagation in foundation models by introducing a localized counterfactual generation method that modifies only attribute-relevant regions in images, resulting in higher visual and semantic fidelity than state-of-the-art methods and measurable bias reduction, such as a decrease in gender classification disparity and balanced person preference scores, while preserving ImageNet zero-shot performance.
Foundation models trained on web-scraped datasets propagate societal biases to downstream tasks. While counterfactual generation enables bias analysis, existing methods introduce artifacts by modifying contextual elements like clothing and background. We present a localized counterfactual generation method that preserves image context by constraining counterfactual modifications to specific attribute-relevant regions through automated masking and guided inpainting. When applied to the Conceptual Captions dataset for creating gender counterfactuals, our method results in higher visual and semantic fidelity than state-of-the-art alternatives, while maintaining the performance of models trained using only real data on non-human-centric tasks. Models fine-tuned with our counterfactuals demonstrate measurable bias reduction across multiple metrics, including a decrease in gender classification disparity and balanced person preference scores, while preserving ImageNet zero-shot performance. The results establish a framework for creating balanced datasets that enable both accurate bias profiling and effective mitigation.