Cross-Loss Influence Functions to Explain Deep Network Representations

TL;DR

This paper extends influence functions to unsupervised and semi-supervised deep learning, enabling model explainability and bias detection in settings where training and testing objectives differ.

Contribution

We introduce the first theoretical and empirical method for estimating influence in cross-loss settings, broadening explainability tools beyond supervised learning.

Findings

Cross-loss influence estimates outperform traditional methods.

Method enables explanation of cluster memberships.

Identifies and mitigates biases in language models.

Abstract

As machine learning is increasingly deployed in the real world, it is paramount that we develop the tools necessary to analyze the decision-making of the models we train and deploy to end-users. Recently, researchers have shown that influence functions, a statistical measure of sample impact, can approximate the effects of training samples on classification accuracy for deep neural networks. However, this prior work only applies to supervised learning, where training and testing share an objective function. No approaches currently exist for estimating the influence of unsupervised training examples for deep learning models. To bring explainability to unsupervised and semi-supervised training regimes, we derive the first theoretical and empirical demonstration that influence functions can be extended to handle mismatched training and testing (i.e., "cross-loss") settings. Our formulation enables us to compute the influence in an unsupervised learning setup, explain cluster memberships, and identify and augment biases in language models. Our experiments show that our cross-loss influence estimates even exceed matched-objective influence estimation relative to ground-truth sample impact.