On the Performance of Concept Probing: The Influence of the Data (Extended Version)

TL;DR

This paper investigates how the choice and amount of data influence the effectiveness of concept probing in interpreting neural networks, specifically in image classification, and provides new concept labels for popular datasets.

Contribution

It explores the impact of training data on concept probing performance and offers publicly available concept labels for key datasets.

Findings

Data quality and quantity significantly affect probing accuracy

Probing models trained on more representative data perform better

New concept labels for standard datasets are introduced

Abstract

Concept probing has recently garnered increasing interest as a way to help interpret artificial neural networks, dealing both with their typically large size and their subsymbolic nature, which ultimately renders them unfeasible for direct human interpretation. Concept probing works by training additional classifiers to map the internal representations of a model into human-defined concepts of interest, thus allowing humans to peek inside artificial neural networks. Research on concept probing has mainly focused on the model being probed or the probing model itself, paying limited attention to the data required to train such probing models. In this paper, we address this gap. Focusing on concept probing in the context of image classification tasks, we investigate the effect of the data used to train probing models on their performance. We also make available concept labels for two widely used datasets.