Delving into Semantic Scale Imbalance

TL;DR

This paper introduces the concept of semantic scale imbalance to better understand model bias in long-tailed data, proposing a new measurement and training framework that improves performance across diverse datasets.

Contribution

It defines and quantifies semantic scale imbalance, and develops a semantic-scale-balanced learning method that enhances model performance on various datasets.

Findings

Semantic scale correlates with classification performance.

The proposed method improves results on long-tailed datasets.

Model bias persists even with balanced data, explained by semantic scale imbalance.

Abstract

Model bias triggered by long-tailed data has been widely studied. However, measure based on the number of samples cannot explicate three phenomena simultaneously: (1) Given enough data, the classification performance gain is marginal with additional samples. (2) Classification performance decays precipitously as the number of training samples decreases when there is insufficient data. (3) Model trained on sample-balanced datasets still has different biases for different classes. In this work, we define and quantify the semantic scale of classes, which is used to measure the feature diversity of classes. It is exciting to find experimentally that there is a marginal effect of semantic scale, which perfectly describes the first two phenomena. Further, the quantitative measurement of semantic scale imbalance is proposed, which can accurately reflect model bias on multiple datasets, even on sample-balanced data, revealing a novel perspective for the study of class imbalance. Due to the prevalence of semantic scale imbalance, we propose semantic-scale-balanced learning, including a general loss improvement scheme and a dynamic re-weighting training framework that overcomes the challenge of calculating semantic scales in real-time during iterations. Comprehensive experiments show that dynamic semantic-scale-balanced learning consistently enables the model to perform superiorly on large-scale long-tailed and non-long-tailed natural and medical datasets, which is a good starting point for mitigating the prevalent but unnoticed model bias.