Encode Bias

Encode bias refers to the unintentional incorporation of societal prejudices into machine learning models, leading to unfair or discriminatory outcomes. Current research focuses on identifying and mitigating these biases across various model types, including transformers, diffusion models, and collaborative filtering algorithms, employing techniques like adversarial training, iterative gradient-based projection, and contrastive learning to improve fairness while preserving model performance. This work is crucial for ensuring the ethical and responsible development of AI systems, impacting fields ranging from image generation and natural language processing to recommendation systems and computer vision.

Papers