Influence Tuning: Demoting Spurious Correlations via Instance Attribution and Instance-Driven Updates
This addresses the issue of spurious correlations in NLP models for researchers and practitioners, though it appears incremental as it builds on prior interpretation work.
The paper tackles the problem of deep learning NLP models relying on spurious correlations by proposing influence tuning, a method that uses model interpretations to update parameters away from spurious patterns, and shows it significantly outperforms adversarial training baselines in deconfounding models.
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.