Adversarial Filters of Dataset Biases

Abstract

Large neural models have demonstrated human-level performance on language and vision benchmarks, while their performance degrades considerably on adversarial or out-of-distribution samples. This raises the question of whether these models have learned to solve a dataset rather than the underlying task by overfitting to spurious dataset biases. We investigate one recently proposed approach, AFLite, which adversarially filters such dataset biases, as a means to mitigate the prevalent overestimation of machine performance. We provide a theoretical understanding for AFLite, by situating it in the generalized framework for optimum bias reduction. We present extensive supporting evidence that AFLite is broadly applicable for reduction of measurable dataset biases, and that models trained on the filtered datasets yield better generalization to out-of-distribution tasks. Finally, filtering results in a large drop in model performance (e.g., from 92% to 62% for SNLI), while human performance still remains high. Our work thus shows that such filtered datasets can pose new research challenges for robust generalization by serving as upgraded benchmarks.

arXiv commentAccepted to ICML 2020

Cited by in corpus (22)

• Measuring Robustness to Natural Distribution Shifts in Image Classification
• Generative Data Augmentation for Commonsense Reasoning
• Adversarial NLI: A New Benchmark for Natural Language Understanding
• Data and its (dis)contents: A survey of dataset development and use in machine learning research
• FINE Samples for Learning with Noisy Labels
• Challenges in Automated Debiasing for Toxic Language Detection
• UnbiasedNets: A Dataset Diversification Framework for Robustness Bias Alleviation in Neural Networks
• Latent Adversarial Debiasing: Mitigating Collider Bias in Deep Neural Networks
• A Survey on Out-of-Distribution Evaluation of Neural NLP Models
• DQI: Measuring Data Quality in NLP
• Our Evaluation Metric Needs an Update to Encourage Generalization
• An Empirical Study on Model-agnostic Debiasing Strategies for Robust Natural Language Inference
• Improving QA Generalization by Concurrent Modeling of Multiple Biases
• TIMEDIAL: Temporal Commonsense Reasoning in Dialog
• Paired Examples as Indirect Supervision in Latent Decision Models
• Competency Problems: On Finding and Removing Artifacts in Language Data
• Mind Your Outliers! Investigating the Negative Impact of Outliers on Active Learning for Visual Question Answering
• COM2SENSE: A Commonsense Reasoning Benchmark with Complementary Sentences
• Combating Unknown Bias with Effective Bias-Conflicting Scoring and Gradient Alignment
• How Does Counterfactually Augmented Data Impact Models for Social Computing Constructs?
• Investigating the Effect of Natural Language Explanations on Out-of-Distribution Generalization in Few-shot NLI
• Coreference Reasoning in Machine Reading Comprehension