Influence Tuning: Demoting Spurious Correlations via Instance Attribution and Instance-Driven Updates

Abstract

Among the most critical limitations of deep learning NLP models are their lack of interpretability, and their reliance on spurious correlations. Prior work proposed various approaches to interpreting the black-box models to unveil the spurious correlations, but the research was primarily used in human-computer interaction scenarios. It still remains underexplored whether or how such model interpretations can be used to automatically "unlearn" confounding features. In this work, we propose influence tuning--a procedure that leverages model interpretations to update the model parameters towards a plausible interpretation (rather than an interpretation that relies on spurious patterns in the data) in addition to learning to predict the task labels. We show that in a controlled setup, influence tuning can help deconfounding the model from spurious patterns in data, significantly outperforming baseline methods that use adversarial training.

arXiv commentFindings of EMNLP 2021

References in corpus (9)

• Understanding Black-box Predictions via Influence Functions
• Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
• Estimating Training Data Influence by Tracing Gradient Descent
• Representer Point Selection for Explaining Deep Neural Networks
• Deep Weighted Averaging Classifiers
• Fine-grained Sentiment Analysis with Faithful Attention
• Input Similarity from the Neural Network Perspective
• Feature-Wise Bias Amplification
• SelfExplain: A Self-Explaining Architecture for Neural Text Classifiers