Soumya Barikeri

RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models

Text representation models are prone to exhibit a range of societal biases, reflecting the non-controlled and biased nature of the underlying pretraining data, which consequently leads to severe ethical issues and even bias amplification. Recent work has predominantly focused on measuring and mitiga…