RepMatch: Quantifying Cross-Instance Similarities in Representation Space

TL;DR

RepMatch is a new method that measures the similarity between groups of training data by analyzing the knowledge encoded in models trained on those groups, enabling broader dataset comparisons and insights.

Contribution

It introduces a novel similarity-based framework for dataset analysis that extends beyond individual instances to compare arbitrary data subsets across datasets.

Findings

RepMatch effectively compares datasets and identifies representative subsets.

It uncovers heuristics behind challenge dataset construction.

The method improves understanding of dataset structure and quality.

Abstract

Advances in dataset analysis techniques have enabled more sophisticated approaches to analyzing and characterizing training data instances, often categorizing data based on attributes such as ``difficulty''. In this work, we introduce RepMatch, a novel method that characterizes data through the lens of similarity. RepMatch quantifies the similarity between subsets of training instances by comparing the knowledge encoded in models trained on them, overcoming the limitations of existing analysis methods that focus solely on individual instances and are restricted to within-dataset analysis. Our framework allows for a broader evaluation, enabling similarity comparisons across arbitrary subsets of instances, supporting both dataset-to-dataset and instance-to-dataset analyses. We validate the effectiveness of RepMatch across multiple NLP tasks, datasets, and models. Through extensive experimentation, we demonstrate that RepMatch can effectively compare datasets, identify more representative subsets of a dataset (that lead to better performance than randomly selected subsets of equivalent size), and uncover heuristics underlying the construction of some challenge datasets.