Predicting and Enhancing the Fairness of DNNs With the Curvature of Perceptual Manifolds
- 27 January 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 47 (5) , 3394-3411
Abstract
To address the challenges of long-tailed classification, researchers have proposed several approaches to reduce model bias, most of which assume that classes with few samples are weak classes. However, recent studies have shown that tail classes are not always hard to learn, and model bias has been observed on sample-balanced datasets, suggesting the existence of other factors that affect model bias. In this work, we first establish a geometric perspective for analyzing model fairness and then systematically propose a series of geometric measurements for perceptual manifolds in deep neural networks. Subsequently, we comprehensively explore the effect of the geometric characteristics of perceptual manifolds on classification difficulty and how learning shapes the geometric characteristics of perceptual manifolds. An unanticipated finding is that the correlation between the class accuracy and the separation degree of perceptual manifolds gradually decreases during training, while the negative correlation with the curvature gradually increases, implying that curvature imbalance leads to model bias. We thoroughly validate this finding across multiple networks and datasets, providing a solid experimental foundation for future research. We also investigate the convergence consistency between the loss function and curvature imbalance, demonstrating the lack of curvature constraints in existing optimization objectives. Building upon these observations, we propose curvature regularization to facilitate the model to learn curvature-balanced and flatter perceptual manifolds. Evaluations on multiple long-tailed and non-long-tailed datasets show the excellent performance and exciting generality of our approach, especially in achieving significant performance improvements based on current state-of-the-art techniques. Our work opens up a geometric analysis perspective on model bias and reminds researchers to pay attention to model bias on non-long-tailed and even sample-balanced datasets.Keywords
All Related Versions
Funding Information
- Key Scientific Technological Innovation Research Project of the Ministry of Education
- National Natural Science Foundation of China (U22B2054)
- Key Project of National Natural Science Foundation of China (62431020)
- National Natural Science Foundation of China (62076192, 62276199, 62276201)
- Higher Education Discipline Innovation Project
- Program for Cheung Kong Scholars and Innovative Research Team in University (IRT 15R53)
- Chinese Ministry of Education
- National Key Laboratory of Human-Machine Hybrid Augmented Intelligence
- Xi'an Jiaotong University (HMHAI-202404, HMHAI-202405)
This publication has 58 references indexed in Scilit:
- Geometric Prior Guided Feature Representation Learning for Long-Tailed ClassificationInternational Journal of Computer Vision, 2024
- Class-Difficulty Based Methods for Long-Tailed Visual RecognitionInternational Journal of Computer Vision, 2022
- Class-Wise Difficulty-Balanced Loss for Solving Class-Imbalance2021
- The Devil Is in Classification: A Simple Framework for Long-Tail Instance Segmentation2020
- Learning From Multiple Experts: Self-paced Knowledge Distillation for Long-Tailed Classification2020
- Feature Space Augmentation for Long-Tailed Data2020
- A Geometric Understanding of Deep LearningEngineering, 2020
- Linear Algebra and Optimization for Machine Learning2020
- Empirical Risk Minimization2017
- ImageNet Large Scale Visual Recognition ChallengeInternational Journal of Computer Vision, 2015