Guide the Learner: Controlling Product of Experts Debiasing Method Based on Token Attribution Similarities
This addresses the issue of dataset biases affecting model robustness for NLP practitioners, though it is incremental as it builds on existing debiasing methods.
The paper tackles the problem of improving out-of-distribution (OOD) performance in NLP tasks by mitigating dataset biases, introducing a fine-tuning strategy that uses token attribution similarities in a Product of Experts loss, which improves OOD results while maintaining in-distribution performance on benchmarks like natural language inference and fact verification.
Several proposals have been put forward in recent years for improving out-of-distribution (OOD) performance through mitigating dataset biases. A popular workaround is to train a robust model by re-weighting training examples based on a secondary biased model. Here, the underlying assumption is that the biased model resorts to shortcut features. Hence, those training examples that are correctly predicted by the biased model are flagged as being biased and are down-weighted during the training of the main model. However, assessing the importance of an instance merely based on the predictions of the biased model may be too naive. It is possible that the prediction of the main model can be derived from another decision-making process that is distinct from the behavior of the biased model. To circumvent this, we introduce a fine-tuning strategy that incorporates the similarity between the main and biased model attribution scores in a Product of Experts (PoE) loss function to further improve OOD performance. With experiments conducted on natural language inference and fact verification benchmarks, we show that our method improves OOD results while maintaining in-distribution (ID) performance.