What Makes a Good Example? Modeling Exemplar Selection with Neural Network Representations

TL;DR

This paper investigates how neural network representations can model human exemplar selection, revealing that strategies emphasizing joint representativeness and diversity best align with human teaching behavior, especially with transformer models.

Contribution

It introduces a neural network-based framework for modeling human exemplar selection, emphasizing joint representativeness and diversity, and compares different neural architectures in this context.

Findings

Strategies based on joint representativeness and diversity best match human judgments.

Transformer-based representations align more closely with human behavior than convolutional networks.

Modeling insights suggest dataset distillation methods can inform machine teaching techniques.

Abstract

Teaching requires distilling a rich category distribution into a small set of informative exemplars. Although prior work shows that humans consider both representativeness and diversity when teaching, the computational principles underlying these tradeoffs remain unclear. We address this gap by modeling human exemplar selection using neural network feature representations and principled subset selection criteria. Novel visual categories were embedded along a one-dimensional morph continuum using pretrained vision models, and selection strategies varied in their emphasis on prototypicality, joint representativeness, and diversity. Adult participants selected one to three exemplars to teach a learner. Model-human comparisons revealed that strategies based on joint representativeness, or its combination with diversity, best captured human judgments, whereas purely prototypical or diversity-based strategies performed worse. Moreover, transformer-based representations consistently aligned more closely with human behavior than convolutional networks. These results highlight the potential utility of dataset distillation methods in machine learning as computational models for teaching.