Revisiting Methods for Finding Influential Examples

TL;DR

This paper critically examines existing methods for identifying influential training examples, revealing their instability and proposing a new evaluation metric based on poisoning attack detection, along with a simple baseline for improvement.

Contribution

It highlights the instability of current influence methods, challenges their evaluation standards, and introduces a new, more effective baseline and evaluation approach.

Findings

Existing influence methods are highly sensitive to initialization and data ordering.

LOO influence and heuristics are poor metrics for explanation quality.

A simple baseline significantly improves influence estimation performance.

Abstract

Several instance-based explainability methods for finding influential training examples for test-time decisions have been proposed recently, including Influence Functions, TraceIn, Representer Point Selection, Grad-Dot, and Grad-Cos. Typically these methods are evaluated using LOO influence (Cook's distance) as a gold standard, or using various heuristics. In this paper, we show that all of the above methods are unstable, i.e., extremely sensitive to initialization, ordering of the training data, and batch size. We suggest that this is a natural consequence of how in the literature, the influence of examples is assumed to be independent of model state and other examples -- and argue it is not. We show that LOO influence and heuristics are, as a result, poor metrics to measure the quality of instance-based explanations, and instead propose to evaluate such explanations by their ability to detect poisoning attacks. Further, we provide a simple, yet effective baseline to improve all of the above methods and show how it leads to very significant improvements on downstream tasks.