Dataset Biases

AbstractState-of-the-art models often make use of superficial patterns in the data that do not generalize well to out-of-domain or adversarial settings. For example, textual entailment models often learn that particular …...

AbstractWe demonstrate the surprising strength of unimodal baselines in multimodal domains, and make concrete recommendations for best practices in future research. Where existing work often compares against random or majority …...

AbstractAlthough achieving promising performance, current Natural Language Understanding models tend to utilize dataset biases instead of learning the intended task, which always leads to performance degradation on out-of-distribution (OOD) samples. …...