How Many and Which Training Points Would Need to be Removed to Flip this Prediction?

TL;DR

This paper introduces efficient influence function-based methods to identify minimal training data subsets whose removal would change a model's prediction, providing insights into model robustness and interpretability.

Contribution

It proposes novel approximation techniques for finding minimal training subsets that flip predictions, enhancing understanding of model robustness and interpretability.

Findings

Influence function-based methods can effectively identify small training sets that flip predictions.

The approach correlates with robustness measures and offers a new way to contest model predictions.

Methods are demonstrated on simple convex text classification models.

Abstract

We consider the problem of identifying a minimal subset of training data $\mathcal{S}_t$ such that if the instances comprising $\mathcal{S}_t$ had been removed prior to training, the categorization of a given test point $x_t$ would have been different. Identifying such a set may be of interest for a few reasons. First, the cardinality of $\mathcal{S}_t$ provides a measure of robustness (if $|\mathcal{S}_t|$ is small for $x_t$, we might be less confident in the corresponding prediction), which we show is correlated with but complementary to predicted probabilities. Second, interrogation of $\mathcal{S}_t$ may provide a novel mechanism for contesting a particular model prediction: If one can make the case that the points in $\mathcal{S}_t$ are wrongly labeled or irrelevant, this may argue for overturning the associated prediction. Identifying $\mathcal{S}_t$ via brute-force is intractable. We propose comparatively fast approximation methods to find $\mathcal{S}_t$ based on influence functions, and find that -- for simple convex text classification models -- these approaches can often successfully identify relatively small sets of training examples which, if removed, would flip the prediction.