FairFlow: Mitigating Dataset Biases through Undecided Learning
This addresses dataset biases in language models, which can improve robustness for AI applications, but it appears incremental as it builds on existing debiasing methods.
The paper tackles the problem of dataset biases in language models, which cause performance drops on new data, by introducing FairFlow, a debiasing framework that uses undecided learning to mitigate biases. The result shows that FairFlow outperforms existing methods, especially on out-of-domain and hard test samples, without compromising in-domain performance.
Language models are prone to dataset biases, known as shortcuts and spurious correlations in data, which often result in performance drop on new data. We present a new debiasing framework called ``FairFlow'' that mitigates dataset biases by learning to be undecided in its predictions for data samples or representations associated with known or unknown biases. The framework introduces two key components: a suite of data and model perturbation operations that generate different biased views of input samples, and a contrastive objective that learns debiased and robust representations from the resulting biased views of samples. Experiments show that FairFlow outperforms existing debiasing methods, particularly against out-of-domain and hard test samples without compromising the in-domain performance